Image fusion is a very important research field in remote sensing. In this paper, we propose a fusion algorithm that combines PCA and discrete wavelet transform (DWT). The first principal component of multispectral images and panchromatic image is decomposed by DWT, the focus in the fusion procedure is choosing a set of basic operations on particular sets of wavelet coefficients which correspond to certain frequency bands. Since previous researches had considered relatively simple fusion rules for combining the wavelet coefficients, we developed a multiple criteria based image fusion algorithm which combines several fusion rules including pixel-level rules and feature-level rules to make the final fusion decision under the general framework for multiresolution image fusion. The experiment result of fusion IKONOS panchromatic and multispectral images shows that the proposed method can improve spatial resolution of multispectral images and is suitable for human vision system observation.

The WTMM-based descriptors, unlike global contour descriptors such as the Fourier descriptors, provide precise local sharp information. This paper introduces an algorithm of image enhancement based on modulus maxima for ground-penetrating radar (GPR) targets detection. By tracking the wavelet transform local maxima at fine scales, singularities can be detected, and the important information of image is also selected. According to the singularities of GPR A-scan signal and the characteristic of GPR signal, GPR image can be commendably given prominence to the targets. Finally, the enhancement of image based on modulus maxima was successfully realized.

Terra / MODIS has spectral and spatial resolution advantage over NOAA/AVHRR. A method of automatically extracting winter wheat area planted was set up applying spectral features with overall remote sensing images. In the paper, extracting winter wheat area planted in Beijing China was taken as example. Firstly, according to winter wheat phonological stage, the best time phase of extraction is in the sowing stage and tillering stage. And the two time phases MODIS data are produced in Beijing China. Secondly, according to the spectral and biological characteristic of the crop, the spectral reflectances of MODIS are analyzed. Red, blue, NIR and ESWIR band were selected as working bands, and the enhanced vegetation index (EVI), which is defined by Red, NIR and Blue bands, and EVI21, which is defined by the difference EVI of the two time phase images were used to improve the precision. Finally, ISODATA algorithm in clustering was used with unsupervised classification. To verify the result, Landsat ETM+ was used to verify its precision. The result shows that the precision reaches 96%. This shows that it can obviously improve accuracy of extracting crop area planted with spectral features, and especially the extraction time can be advanced for more than three months compared with traditional method, which thought the best extracting winter wheat area planted time was in March. So the method of spectral features for extracting area planted can fine to meet the operating method for agricultural condition monitoring with remote sensing and information service system.

Over the last few years, we have developed an algorithm which detects anomalous targets in hyperspectral images. The algorithm takes a hyperspectral cube with a completely unknown background, segments the cube, assigns the largest clusters as background, and determines which pixels are anomalous to the background. In the work to be presented here, we will add two additional modules. First, since our present mission is to detect military targets in a fairly barren rural background, we use the SAVI (or NDVI) metric to detect items which appear to contain chlorophyll. In this way, we can eliminate objects which in retrospect were the right sizes and shapes but were in reality plants. Second, we have developed CFAR methods to achieve a Constant False Alarm Rate while giving us the maximum probability of detecting the targets. Actual data will be analyzed by the algorithm; the ability to both determine if a target is present and where its location is will be shown.

It is one of basis of water color remote sensing to investigate the method to obtain and analyze the spectral features of the water bodies. This paper concerns the above-water method for the spectral measurements of inland water. A series of experiments were taken in areas of the East Lake with the EPP2000CCD radiometer, and the geometry attitude of the observation and the method of the elimination of the noise of the water Signals will be discussed. The method of the above-water spectral measurements was studied from the point of view of error source. On the basis of the experiments of the water depth and the observing direction form the sun and surface, it is suggested to remove the radiances of the whitecaps, surface-reflected sun glint and skylight which have not the spectral features of water from the lake surface by specialized observing attitude and data processing. At last, a suit of methods is concluded for the water body of the East Lake in measuring and analyzing the spectral features from above-water.

The sea surface as an important form of background usually plays a decisive role in sea applications. Real backgrounds are expensive while sea simulation can generate realistic images under various conditions. In this paper, a physics based model of infrared image synthesis for wind-ruffled sea is proposed. It consists of an integrated process based on oceanographic models of sea waves and radiometric properties of sky and sea. First, an analytical, semi-empirical model is introduced to construct height fields as an approximation of the sea surface based on Gaussian random numbers with spatial spectra of a prescribed form. Then, an improved model is constructed to calculate the radiance of the water surface. Infrared sea radiance is therefore composed of thermal self-emission of the sea surface, reflected radiance of the sun (moon), sky and clouds, and radiance of the atmospheric path between the sensor and the sea surface. An infrared imaging illumination model is proposed to render the sea surface based on Torrance-Sparrow reflectance model. The radiance is calculated by ray tracing. Additionally, infrared effective emissivity of sea surface is incorporated to enhance the phenomenal realism. Finally, some results of real-time animation of wind-ruffled sea surface in atmospheric window are shown. The generating images of the sea surface are in good accordance with real images.

Site-specific variable nitrogen application is one of the major precision crop production management operations. To obtain sufficient crop nitrogen stress information is essential for achieving effective site-specific nitrogen applications. This paper presents the development of a multi-spectral nitrogen deficiency sensor, which uses three channels (green, red, near-infrared) of crop images to determine nitrogen level of the rape. This sensor assess the nitrogen stress by means of the estimated SPAD value of the rape based on rape canopy reflectance sensed using three channels (green, red, near-infrared) of the multispectral camera. The core of this investigation is the calibration methods between the multi-spectral references and the nitrogen levels in crops measured using a SPAD 502 chlorophyll meter. Based on the results obtained from this study (The correlation was 0.89.), it can be concluded that a multi-spectral CCD camera can provide sufficient information to perform reasonable SPAD values estimation on-the-go during field operations.

Fusion is basically extraction of best of inputs and conveying it to the output. In this paper, we present an image fusion technique using the concept of perceptual information across the bands. This algorithm is relevant to visual sensitivity and tested by merging multisensor, multispectral and Defoucused images. Fusion is achieved through the formation of one fused pyramid using the DWT coefficients from the decomposed pyramids of the source images. The fused image is obtained through conventional discrete wavelet transform (DWT) reconstruction process. Results obtained using the proposed method show a significant reduction of distortion artifacts and a large preservation of spectral information.

NOAA AVHRR/3 data are very important in scientific study and operational applications, which have been widely used in environment monitoring, drought monitoring and crop yield evaluation in China. For this reason, it is necessary to set up a stable, robust and automatic data preprocessing system. In this paper, a NOAA AVHRR/3 data preprocessing System is designed and implemented, which is developed by Interactive Data Language (IDL). At first in the paper, the importance of The NOAA AVHRR/3 data preprocessing is addressed. And then an architecture of NOAA AVHRR/3 data Preprocessing System is presented, which includes the Automatic Data Receiving, Radiant Calibration, Brightness Temperature Calibration, Geometric Correction, Cloud Detect, Atmosphere Correction, BRDF Correction, Land Surface Parameter Retrieval, Data Visualization and Mapping. In addition, Management of above output data is realized using Oracle 9.0 in this paper. At last, the performance of this system is analyzed .The result shows that this architecture of NOAA AVHRR/3 Data Preprocessing System is feasible.

Leaf Area Index (LAI), as a fundamental parameter to evaluate the physiological condition of plants, was calculated by image processing based on machine vision technology. The measurement system hardware consisted primarily of the MS3100 3CCD camera, the image grabber card, a desktop computer and the acquired images were processed by Matlab and ENVI. After acquiring the 3 images by the 3CCD camera of Green, Red and NIR channels, the NIR image was considered more effective in separating the soil background for its higher contrast value. Thus, it was selected for image processing to calculate the leaf area index (LAI). The transect method was applied to obtain the threshold 50 in the binary image conversion and the soil background was thus eliminated as a result that most of its reflectance in the image was under 50. Then the 'imerode'-'imdilate' operation in the image processing box of Matlab was used to remove the left crop stem noises, including those small weeds in the binary image background. Consequently, the LAI of the acquired NIR image was calculated as 0.523 by dividing the total image pixel amount by that of the black pixels in the binary image.

A system for detection of the dim moving small targets in the IR image sequences is presented. In the paper, a new algorithm was proposed. It has high performance for detecting moving small targets in infrared image sequences containing cloud clutter. One advantage of this approach of detecting dim small targets is that it fuses the features of the moving small targets in both the spatial domain and temporal domain. Two independent units can realize the calculative process. The efficiency of the algorithm is tested by two kinds of representative IR image sequences sampled from the real environment. The results of it show that the algorithm we presented has high ratio of detection precision.

Efficient VLSI architectures for multi-dimensional (m-D) discrete wavelet transform (DWT), e.g. m=2, 3, are presented, in which the lifting scheme of DWT is used to reduce efficiently hardware complexity. The parallelism of 2m subbands transforms in lifting-based m-D DWT is explored, which increases efficiently the throughput rate of separable m-D DWT. The proposed architecture is composed of m2m-1 1-D DWT modules working in parallel and pipelined, which is designed to process 2m input samples per clock cycle, and generate 2m subbands coefficients synchronously. The total time of computing one level of decomposition for a 2-D image (3-D image sequence) of size N2 (MN2) is approximately N2/4 (MN2/8) intra- clock cycles (ccs). An efficient line-based architecture framework for both 2D+t and t+2D 3-D DWT is first proposed. Compared with the similar works reported in previous literature, the proposed architecture has good performance in terms of production of computation time and hardware cost. The proposed architecture is simple, regular, scalable and well suited for VLSI implementation.

Snow is highly reflective and snow albedo play an important role in surface energy fluxes. Snow albedo is an important input parameter needed by both global and regional climateic models. In addition the albedo of snow-covered surfaces varies during relatively short time periods. The Tibetan Plateau is the most sensitive area in the world to hydrological cycle and climatic change. Traditional in situ surveying and mapping cannot provide enough large-scale snow cover information; in addition it will take much time and money, sometimes even dangerous. Remote sensing is a practical means for mapping snow albedo in high elevation area such as in the Tibetan Plateau. The conventional methods estimate snow albedo through a series of processing steps including atmospheric correction, surface angular modeling, and narrow band to broadband albedo conversions. The accuracy of estimating snow albedo relied on all these processes. This paper develops a direct method to estimate snow broadband albedo using Moderate Resolution Imaging Spectroradiometer (MODIS) imagery in Tibetan Plateau.

In-site experiments are conducted to simulate a group of wheat colonies with gradually decreasing Leaf Area Index (LAI) by removing individual wheat plant from an original wheat field. The LAIs vary from the original 5.2 (sufficient saturation) to 0 (uncovered soil) at the step of 0.52 or 10% of the original one. The potential and reliability of using remote sensing method for crop growing assessment was discussed. Taking the curve of uncovered soil's reflectance spectra as a reference line, and using the distance between the curve of the reflectance spectra at a specific LAI level and the reference line as a measure, at the locations of three feature spectra in visible and infrared band, i.e., 680nm, 890nm and 1460nm, the response of reflectance spectra of wheat colony to the variance of its LAI was studied. The results show that with the reduction of LAI, the maximal closing motion of wheat colony curve to the soil reference line occurred at 890nm, followed by 1460nm and 680nm. And when LAIs decrease gradually from 5.2 (100 percent of wheat colony) to 4.68(90 percent colony), to 4.16 (80 percent colony), and finally to 0, the approaching speeds of wheat colony spectral curves to soil spectral curve are different. In the range of 5.2(100%) to 2.6 (50%), the approaching speed is smoothly low, especially in the range of 4.68 (90%) to 2.6(50%), while the closing speed in the range of 2.08 (40%) to 0.52 (10%) mount to very high.

We propose a deblocking algorithm using multi-scale edge representation of images. All block-based image/video coding methods suffer the annoying blocking artifacts at low bit rate. Blockings are considered as tiny edges occur at the borders of coding blocks in the directions of horizontal or vertical. These directional and magnitude specialties characterize blockings from other common edges in nature images, and can be reasonably used in the designing of deblocking algorithms. We decompose an image using dyadic wavelet transform and then find the local extrema of wavelet coefficients, which are corresponding to edge points. The extrema caused by blockings are eliminated using some predefined criteria. The de-blocked image can be reconstructed from those filtered extrema with a Projection Onto Convex Sets (POCS) method. Experimental results show that the proposed method can generate fairly well deblocking results under various bit rate conditions.

The universal pattern decomposition method (UPDM) has been successfully applied to simulated data for Landsat/ETM+, Terra/MODIS, ADEOS-II/GLI and 92 bands-CONTINUE sensors using ground-measured data. This paper validates the UPDM using MODIS and ETM+ data acquired over the Three Gorges region of China. The reduced 2 values of selected area D, that with the smallest terrain influences, are 0.000409 (MODIS) and 0.000181 (ETM+), and the average linear regression factor between MODIS and ETM+ is 1.0077, with rms 0.0082. The results demonstrated that the UPDM coefficients are sensor-independent.

Shuttle Radar Terrain Mission (SRTM) DEM has become one of digital topographic data sources of the earth because of its high spatial resolution and near-global coverage. However, its widely usage has been limited by some void areas occurred in SRTM DEM, which are mainly related to the water body, spikes or wells. Although they were modified into finished SRTM DEM by using a complicated process by National Geospatial-Intelligence Agency (NGA), in which a lot of void areas could be filled with correct data, some void areas still exited especially in the water area. In addition, the accuracy of the finished SRTM DEM might be hindered because of no global accurate DEM as a reference. And the finished SRTM DEM can't be freely downloaded from Internet also limits its usage in some extent. A simple method to create an edited SRTM DEM based-on Landsat ETM+ image was proposed in this paper. The unfinished SRTM data was firstly re-projected to the UTM projection for matching the Landsat ETM+ image in the same area. Secondly, through analyzing the spectral attributes of water, the water body was accurately extracted from Landsat ETM+ image by using the indices of NDWI and NDVI. Thirdly, the water body in the unfinished SRTM DEM was masked, and the water void areas and non-water void areas were finally separated. The water body void areas were filled with the surrounding minimum elevation data and the non-water void areas were filled by using the method of Kriging interpolation. The results showed that the proposed method could improve the unfinished SRTM DEM, which were proved to be better than CIAT edited SRTM DEM according to the comparison of both visual effect and statistical accuracy.

Digital watermarking is the technique which embeds an invisible signal including owner identification and copy control information into multimedia data such as image, audio and video for copyright protection. A blind robust algorithm of hiding data into H.264 video stream rapidly was proposed in this paper, copyright protection can be achieved by embedding the robust watermark during the procedure of intra prediction encoding which is characteristic of H.264 standard. This scheme is well compatible with H.264 video coding standard and can directly extract the embedded data from the watermarked H.264 compression video stream without using the original video. Experimental results demonstrate that this scheme is very computational efficient during watermark embedding and extraction and the embedded data not lead to increasing the bit-rate of H.264 bit-stream too many. This algorithm is feasible for real time system implementation.

Traditional remote sensing image classification methods have been mature,especially the maximum likelihood technique based on statistical analysis methods. But Traditional remote sensing image classification methods can't handle multiple source remotely sensed data,in order to make optimized decisions, better use must be made of all available information acquired from different sources.Evidential reasoning has been proposed as one of the most promising approaches for integrating multisource information. We expatiate on Dempster-Shafer evidence theory and present a method of multisource information fusion based on it. We also apply this method in urban development. The experimental results show that this method present in this paper is effective,and it can greatly improve the ability of image classification.

This paper describes an algorithm framework for registration of laser scanned range data and multiple overlap images. In our approach linear features were used for pose estimation. Many conjugate lines are automatically generated by extraction of linear primitives from images and laser data. The exterior orientation parameters of the images are calculated based on the theory and the arithmetic of the line photogrammetry, in which the conjugate lines are used as the observation values. The coplanar condition was used as error equation to resolve the external orientation parameter of digital image. In addition to the theoretical method, the paper presents a experimental analysis based on the quality of fit of final alignment between the LIDAR and digital cameras.

Spectral Angle Mapper (SAM) model has got wide applications in hyperspectral Remote Sensing (RS) information processing. But Spectral Angle couldn't achieve satisfied performance in some cases because of its sensitivity to noises and uncertainty. Based on the analysis to traditional SAM algorithm, four types of errors and their impacts to spectral angle are investigated. In order to reduce the impacts of above errors, some improved algorithms are proposed and experimented. The first improved algorithm is grouping spectral angle algorithm. In this new algorithm all bands are divided into two sets by odd and even bands, that means two additional sub-vectors are created in addition to the original spectral vector. So three spectral angles will be computed and the minimum of three indexes is used as final index. The second improved algorithm is normalized spectral angle. In this way spectral angle is computed to the normalized vectors of two original vectors. Two approaches are used to normalize the spectral vector, and spectral angle is computed to the normalized vectors. This algorithm is able to decrease the impacts of random errors. The third algorithm is intersected spectral angle. Spectral angle is calculated by a spectral displacement strategy in this approach. That means a given displacement to change the corresponding bands of two spectral vectors is used and a spectral angle to the displaced vectors will be got. By this displacement strategy the impacts of band offset is reduced. Finally some experiments are used to test those improved algorithms. It proves that those new approaches can reduce and control the errors and improve the precision and reliability of similarity measure.

The purpose of this paper is to clarify the components of surface water spectral energy as well as the factors that affect these components. Using proper spectral measurement technology and data retrieval algorithms, the spectral characteristic of varying water quality can be obtained, based on the fact that inland water spectral reflectance curves are generally different because of the differences caused by the organism, the eutrophication, and the suspended sediment content(SSC). A model was developed based on the ratio of the water chlorophyll a concentration to spectrum, with a normalized model taking care of simultaneously observed data concerning water quality parameters. The relationship between the pollution index of East Lake in Wuhan and spectral reflectance characteristic was also described through a mixed model concerning chlorophyll and sediment reflectance. The models were proved by some experiments, which facilitate the inversion of water pollution index based on remote sensor measurement.

The appearance of imaging spectrometer enables us to acquire hyperspectral remote sensing images, so we can acquire abundant spectral information about earth material, and we can analyse mixed pixels questions. In this paper, a hyperspectral image about a region given is analysed. We unmix the earth material interested from the hyperspectral image about the region by using Convex Geometry Analysis(CGA) method. The CLS proportion images and inherent proportion images about earth material interested of the hyperspectral image are obtained.

Detecting unkown man-made targets in an unknown background is a great challenge in hyperspectral imagery analysis since all of the prior knowledge about targets, backgrouds and noise is not available. In this paper, we present an automatic spectral detection algorithm to deal with the problem. Unlike some hyperspectral target detection algorithm which take advantage of the prior spectral signature, the proposed algorithm is to estimate the spectral signaure completely from the observation and removing undesired signature using linear spectral mixture model and subspace projection before feature-level fusion. It consists of several successive processes: (1)estimating the spectral signatures of background and targets using eigenvalue analysis and automatic target detection and classification algorithm (ATDCA); (2)decomposing the observation space into a noise space and a signature space spaned by target and background spectral signatures; (3)projecting hyperspectral datum onto the signature subspace in order to reduce the noise effects; (4)projecting residual datum onto orthogonal complement subspace of background space spaned by backgroud spectral signatures and onto subspace spaned by targets spectral signatures, thus suppressing the residual undesired spectral signatures; and (5)a generalized likelihood ratio test(GLRT) which, as a fusion center, is used to achieve detection output from component images at feature level. The algorithm is tested with a HYDICE hyperspectral imagery in which simulated targets have been implanted. The results of experiment and theoretic analysis verify the effectiveness of the algorithm.

Building extraction from airborne LIDAR data has gained much focus in recent year. Usually, buildings are reconstructed by combination of facets or line frames detected from points cloud. It is difficult to extract buildings with curve shape using these methods. In the paper, we attempt a novel way to extract building by anisotropic surface noise faring and simplification. A slope based building wall detection with isolate surface area size control is used in detect the building boundary. Points of building are taken out within the boundary. After an anisotropic filtering of surface mesh, facets are detected to simplify the model so as to take fewer store volume. Experiment is given to prove the methods in this paper.

Building shadows are very integral parts in the high spatial resolution remote sensing images, especially in urban region. For this reason, shadow detection is a very important step in remote sensing imagery interpretation. Two kinds of method for shadow detection have been developed: color-based and model-based, but the both methods have their disadvantages: the former couldn't differentiate the objects which have similar color property with the shadow; the latter required other data for assistance. An effective approach to building shadow detection in Quickbird multi-spectral imagery is presented in this paper. The method is based on normalized multi-spectral data and object-based classification technology. By analyzing the spectral and geometrical characteristic of shadow region, the procedure is divided into three stages: 1) normalized multi-spectral imagery is obtained, by using 4 single band image data divide the sum of the four band data (similar with the normalized rgb); 2) using the blue and near infrared band of the normalized multi-spectral imagery to determine a threshold to extract shadow candidate regions; 3) an information integration stage confirmed or rejected each detected candidate image object by using object-based classification technology. Experimental results show that the proposed algorithm is efficient in detecting building shadow for Quickbird imagery.

ASTER is a valuable tool in studying and understanding landscape processes at local to regional scales, particularly in China with more fragments of agricultural field, and few studies have been reported on combing its bands to do the Tasseled Cap (K-T) Transformation yet, though this method has been applied to TM data since 1976, so we present an approach for using the Gramm-Schmidt method to perform the Tasseled Cap interactive transformation on ASTER image data. Comparing the results derived from the method with NDVI and PCA, we find that the main surface features, including vegetation with different density, can be distinguished based on their physical characters and the result images accord well with in situ investigation. In conclusion, this method can combine the knowledge about the research site with the information contained and can be reflected by remote sensing. So it makes the process simpler and results more accurate, and can be widely used for feature separation and extraction.

In this paper, a practical and operative method is presented for water quality classification in Lake Tai using multi-spectral remote sensing data. The remote sensing models of chlorophyll-a and suspended sediments are developed by integration of Landsat5 TM images and in situ measurements in the period of from year 2003 to 2004. The key issues in applying multi-spectral satellite imagery to water quality are first given, including radiation calibration, geometric correction, atmospheric correction, noise elimination by spatial filtering and the spectral reflectance obtain. The multi-spectral band analysis is further studied to determine the relations between various bands emissivity, the emissivity combination and the water quality parameters. The regression models for chlorophyll-a and suspended sediments are derived, and the accuracies and correlation coefficients of the models are analyzed. The water quality parameters obtained from the regression models are tested with the in situ measurements, which proves the uniformity between these two results. This means the feasibility of the multi-spectral remote sensing data for monitoring the water quality in Lake Tai.

The color composite DMC images are obtained by calibrating, mosaicking and color composite fusion processing, but the quality usually is still not good enough to be satisfied. That mainly appears that, between the four images in the color composite DMC image, in a narrow range, there is a smooth transition, but for the overall image, the radiometric differences still exist. This paper, based on analyzing the character of the residual radiometric differences in color composite DMC images, proposes a post-processing approach to improve the images' quality and achieve high quality imagery. The procedure employs a multi-scale adjustment scheme to ensure a more satisfied result and each adjustment scheme is performed in a different spatial scale. Finally, experiments are given and results show that the approach proposed by this paper can remove the residual radiometric differences and acquires images with higher quality.

In this paper, we introduced information hiding technique into remote sensing area, proposed its characters, requirements and difference from general information hiding technique and illuminated that general image hiding algorithm doesn't adapt to remote sensing image. We developed the connotation of information hiding that the secret information is still hidden in the original remote sensing image and proposed a practical information hiding technique for multi-spectral remote sensing images and a wavelet information hiding algorithm adapting to the features of the multi-spectral remote sensing image based on DWT embedding strategy and HVS character. The experimental results show that the information hiding technique and algorithm for multi-spectral remote sensing images proposed in the paper not only has the advantages of good transparency, strongness, large information capacity, correct extraction of secret, well protecting the textural and spectral features of the secret ground object and vividly resuming the multi-spectral remote sensing image, but also has a strong robustness against JPEG lossy compression and noise adding. Furthermore this algorithm has no influence on applied value of a multi-spectral remote sensing image, doesn't need the original image while extracting the secret information and resuming the original one and was a blind algorithm.

Anisotropic diffusion has been introduced as an effective tool for image smoothing and extended to multispectral images. In this paper, two improved multispectral anisotropic diffusion approaches are presented. Using a local Gaussian scale selection technique based on probabilistic estimate and the morphological estimate of the gradient threshold, the diffusion approaches are further improved by allowing better control of the diffusion behavior and an appropriate stopping mechanism. The proposed approaches not only effectively remove Gaussian noise and randomly impulsive noise caused by sensors, but also preferably preserve important and detailed edges and image quality. Experimental results on two multispectral satellite images are given to show that the proposed approaches have superiority capability over the previous multispectral anisotropic diffusion schemes on visual judgment and quality statistical analysis.

Multi-spectral images acquired by different image sensors from satellites or aircrafts are often covered with cloud under bad weather condition. In this paper, we propose a new cloud removal method to restore the cloud-covered area of multi-spectral images using only a couple of multi-spectral images taken of the same scene. In our algorithm, we take the registered visual and infrared images as an example. We first de-noise the two images with the method of Wiener Filter to wipe off the primary noise. As the infrared imaging has more powerful ability of penetrating through cloud than the visual image, we adopt the method of Poisson Matting to exactly segment the edge of area covered by infrared cloud and use wavelet analyzing to restore the area originally occupied by infrared cloud. Then a B-spline based model is hired to repair the residual holes. For the corresponding visual image of the same scene by taking the spacial correlation between the two multi-spectral images, the location of targets under visual cloud is reconstructed and their texture styles are finally recovered by colorization method. Our algorithm is easy to manipulate and can also be extended to other multi-spectral wavebands.

The star/galaxy/quasar (qso) classification according to spectra in astronomy is basic and important step in LAMOST project. It is proposed that two algorithms based on the KPCA and the GDA to classify star, galaxy and quasar in this paper. For the performance of classification, the experiments show that the KPCA is slightly better than the PCA; the GDA is also slightly better than the LDA, and it is significantly better than the KPCA.

A new algorithm to enhance contrast for image is proposed, which based on discrete stationary wavelet transform (DSWT) and non-linear gain operator. Comparing with usual discrete orthogonal wavelet transform, DSWT is redundant and shift-invariant. It can give a more approximate estimation to continuous wavelet transform. It can eliminate the "Gibbs" phenomena when image is reconstructed. This will improve greatly quality of reconstructed image. Combining DSWT with generalized cross validation principle, a new de-noising algorithm to image is proposed. The new de-noising algorithm can restrain efficiently white noise and colored noise in the image without prior-knowing variance of noise in the image. An asymptotical optical threshold can be obtained by only data of original image. Having made DSWT to an image, de-noising is done with proposed algorithm in the high frequency sub-bands in the better resolution levels. Contrast is enhanced by combining de-noising algorithm with non-linear gain operator in the high frequency sub-bands in the worse resolution levels. A new criterion to evaluate quality of enhanced image is given. Experimental results show that the new algorithm can suppress white noise and colored noise in the image effectively while it also enhances the contrast of image well. The proposed new algorithm is more excellent in performance than histogram equalization, un-sharpened mask algorithm, WYQ algorithm and GWP algorithm.

Data integration is a very important strategy to obtain optimum solutions in geo-scientific analysis, 3D scene modelling and visualization. This paper mainly focuses on the integration of GIS data, stereo aerial imagery and DSM to derive automatically tree rows and hedges in the open landscape. The roads, field boundaries, rivers and railways from GIS database also represent potential search areas for extracting tree rows and hedges, which are often located parallel and near to them. Different approaches, such as image segmentation by CIE L*a*b, edge extraction, linking, line grouping, space intersection and 3D verifying with DSM, are combined together to extract the objects of interest. The extracted information of tree rows and hedges can be used in many applications, such as deriving of wind erosion risk fields for soil monitoring and protection.

Rough set theory has a powerful capability for attributes reduction and classification rules extraction, while artificial neural network (ANN) performances well in classification problems with a satisfactory accuracy. In this paper we focus our attention to investigate a way of integrating rough set theory and multi layer perceptron (MLP) in soft computing paradigm for classification and rule generation of hyperspectral remote sensing image classification. The novelty of this method lies in applying rough set theory for extracting classification rules and computing fuzzy membership values directly from decision table after attributes reduction on a real-valued attribute table consisting of classification features. The successful application of this approach in hyperspectral remote sensing images mineral classification illustrates the flexibility and practicality of this new approach.

In order to study land cover classification and natural environment, we must analyze vegetation cover states of the local scale in which we can know the subject in detail, as well as and the global scale. Therefore, we need to analyze various satellite data sets which are measured with different wavelengths region and different number of bands. However, it is difficult to compare analysis results obtained using such data sets. By the universal pattern decomposition method (UPDM), which is sensor independent analysis method, we examined vegetation coverage of a pixel on data sets measured different wavelength range and different resolution which is acquired at the same place and time. In this study, in order to develop a generalization rule of vegetation coverage, we examine vegetation coverage of a pixel on 1-kilometer resolution data sets using results obtained by analyzing 250-m resolution data sets which are acquired at the same place and time as the pixel of 1-kilometers'. We defined the rule of classifying into five levels of vegetation coverage using results of high resolution data sets analyzed by the UPDM. Using the results of the analysis, we calculate vegetation coverage of Kii peninsula area.

Algorithms to estimate soil moisture using Advanced Microwave Scanning Radiometer (AMSR) data and experiments to determine their validity have been developed. Since estimations of soil moisture using AMSR data are affected by vegetation moisture content, determination of the quantity and distribution of vegetation is necessary. A variety of information can be obtained simultaneously using optical sensors such as Global Imager (GLI). In this study, we attempted to estimate plant water content using the vegetation index obtained with GLI to determine its sensitivity to vegetation coverage as well as the relationships among vegetation coverage, biomass and plant water content based on field survey data.

Satellite ADEOS-II was launched on 14 December 2002 by Japan Aerospace Exploration Agency (JAXA). The purpose of this study is to estimate global terrestrial net primary production (NPP) with a newly developed algorithm that estimates gross photosynthesis using ADEOS-II/GLI data as input. It is the first time that ADEOS-II/GLI data have been used as input to estimate NPP. The NPP estimation error is 26%. In this study, total ANPP estimates between 60°N and 60°S in the world were 65.2±17.0 [10¹⁵gC/year]. The results were compared with NPP ground measurement data, and NPP values estimated by other studies, such as NPP derived from climatic model and NPP estimated using other satellite data (e.g., NOAA/AVHRR, Terra/MODIS). The pattern of this study's distribution of estimated NPP in the world was similar to that of other studies.

When condition matching used for color representation based on standard observer, metamer exists inherently which lowers the accuracy of color production and color communication. A new multispectral space is developed in this paper by means of hybrid algorithm, i.e., cluster analysis with principal component analysis for color representation. Results show that the multispectral space not only keeping the structure of the original space but eliminating zero-crosses as well its accuracy of color representation is higher.

In this paper, we developed a 3-dimension 1-rank mapping function which is CIELAB space versus the device RGB space based on LMS in RGB subspace, and then implemented color correction of scanning input data by using this function. Experiment proved that the color difference between the original and the computational color data from the individual RGB subspaces with color mapping function is less than that in the whole RGB space.

A method of net primary production (NPP) estimation from pattern-decomposition-based vegetation index (VIPD) using ADEOS-II/GLI data has been developed. But since the global sensor (GLI) could not be directly above the objective when observing, it is necessary to consider the effect of bi-directional reflectance distribution function (BRDF). To validate the method of NPP estimation and for the algorithm for the retrieval of albedo from GLI, bi-directional reflectance factors (BRF) observations for the reflectance of a cedar forest on the Kii peninsula with a sensor onboard an unmanned helicopter were held in July, 2002. In this paper, a kernel-based BRDF model is used to remove the BRDF's effect on the reflectance. The semi-empirical Ross-Li (reciprocal RossThick-LiSparse) model and its performance under conditions of BRF observations are discussed, showing that the retrievals obtained are reliable. The retrieved reflectance is the nadir viewing and the overhead sun could be achieved by this model. And then VIPD could be calculated from the retrieved reflectance. With the data of VIPD and photosynthetically active radiation (PAR), NPP is estimated as 0.36 KgCO₂/m²/month.

Pattern decomposition method (PDM) may be thought to be a type of spectral mixture analysis in which each pixel is expressed as the linear sum of fixed, standard spectral patterns. The usage of fixed standard spectral patterns makes possible the comparison of data from different time and also from different sensors with the same criteria. In the development of the PDM, I introduced new point of view to interpret the PDM. It is multi-dimensional analysis. In a sense, the standard patterns in the PDM are thought as a kind of principal axes in n-dimensional space but have physical meaning. To make it possible, the PDM adopts an oblique coordinate system. The standard patterns are the axes of the coordinate system. There is another mathematical analysis method that uses oblique coordinate system. It is factor analysis method. In factor analysis, there is an algorithm that extracts meaningful factors in oblique coordinate system. In this paper, I apply this algorithm to Landsat/TM data and show the obtained factors are quite similar to the three standard patterns of the PDM.

This paper proposes a new concept of remote sensed image, which could be considered as the mapping mode of the spectral space for ground features to the projective space that is two-dimensional imaging space. This concept could express well the texture information in multi-spectral and super-dimensional images, as well as that in single band image or black-white images. In order to analyze the texture information in multi-spectral and super-dimensional images, the necessary way is the coding for spectral vector of every pixel with all the spectral information. The texture of multi-spectral and super-dimensional images is based upon that kind of coding. Several approaches have been proposed for the coding of spectral vector: the coding based on the spectral similarity, the coding based on the density analysis of spectral space, the coding based on the image analysis of principal components, etc. The result of texture classifications based on one way of coding has been given out to show the effectiveness of proposed method.

Traditionally, suspended sediment concentration (S) has been measured by time-consuming and costly boat surveys which allow the accurate measurement of S for single points in space and time. Remote sensing from spaceborne sensors has proved to be a useful method to such surveys as it provides and instantaneous and synoptic view of sediments that would otherwise be unavailable. The key to success of remote sensing in such a role is to get the suitable relationship between S and remotely sensed spectral radiance (L). In this research, an estimation model of suspended sediments in Yangtze River is built by using spectral analysis based on some correlative researches. Firstly, we attempt to fit the relation curve between suspended sediments concentration and reflectance of water by using three widely used classical models. Land spectral experiment, which simulates the condition of Yangtze River, is carried out to estimate the parameters of three models. And then, we analyze the Landsat TM data by using the models and finally compare the results with the hydrological data for error quantitative analysis.

This paper presents a method for spatiotemporal alignment of two video sequences recorded by aerial sensors using different modularity. It recovers the inter-video temporal synchronization and spatial transformation from two sequences of intra-video transformations between successive frames within each video. Since it needs no directly comparison of images across two videos, and the intra-video transformation is easy to obtain, the alignment is possible even when the two videos have very different appearances. We make best use of the geometry between rigidly connected aerial video sensors and the ground scene, treat the intra-video and inter-video spatial transformations as similar ones, and develop a simple alignment method. Both theoretical analysis and experimental results demonstrate the stability and efficiency advantages over previous methods.

The abrupt shot change detection is a basic and important technology in content-based video retrieval. Some basic algorithms, such as the pixel-matching algorithm, the histogram algorithm, used to detect abrupt shot change in digital video have existed. But existing algorithms can't eliminate the influence of the video movement and were easily affected by the shakes of the shots. When much video movement or shakes existed, existing algorithms can't perform well. In this paper, a new method is proposed, the proposed method consists of two new algorithms: twi-difference algorithm and motion-matching of the potential shots algorithm. The twi-difference algorithm can eliminate the influence of the video movement on the shot change detection. The motion-matching of the potential shots algorithm can reduce the influence of the video shakes. Experimental results have showed that the new method performs well both in recall and precision.

At present there are many methods that could deal well with frontal view face recognition when there is sufficient number of representative training samples. However, few of them can work well when only one training sample per class is available. In this paper, we present a method of face recognition based on wavelet low-frequency band and singular value decomposition (SVD) to solve the one training sample problem. To acquire more information from the single training sample, training image is linearly combined with its reconstructed image of wavelet low-frequency band into a new training image. By using Fourier transform, the spectrum representation of face image is obtained that is invariant against spatial translation. Then the spectrum representation is projected into a uniform eigen-space that is obtained from SVD of standard face image and the coefficient matrix is used as feature for recognition. The proposed algorithm obtains acceptable experimental results on the ORL face database.

The speed of spatial data transmission on the web is too slow to accept in the field of Web 3D visualization. Aiming at this condition, the paper proposes several methods to resolve it. For the network structure, the paper designs a rational architecture between the server and client to achieve the effective load balancing. For the terrain modeling, the authors adopt the Levels of Detail (LOD) technique using the improved binary triangle tree structure. It is similar to ROAM, but easy to be implemented. For the building modeling, the paper discusses the building texture division and compression technique. Finally, experiment results are demonstrated to prove the proposed research methods are really practicable.

The main characters of map generalization are analyzed in this paper. A new buildings' generalization method based on mathematical morphology is proposed to make the algorithm more effective: Firstly vector building objects are transformed into a binary image, in which vector objects are represented by raster features, the binary image is processed by means of morphological algorithms. Small bulged parts of houses are eroded by erosion calculation and small sunken parts of houses are filled by dilation calculation. Lastly the building's edges are extracted and transformed into vector polygons.

To help the walking patient avoid the mobility hazard is one of the key tasks of artificial visual prosthesis: a newly risen research project. A mobility hazard detection algorithm especially for such a system is proposed, in which a U-V-D space model is constructed for the non-hazard targets in the ground plane, so that those objects violating the model are easily identified as obstacles or pits. An iterative maximum likelihood procedure is invented to fit the model accurately and robustly with fast convergence being meanwhile accomplished.

Scene segmentation in digital video plays important role in content-based video indexing. This paper uses the techniques of text detection and audio features extraction, etc, discusses scene segmentation in news video from the point of combination of audio and video, and puts forward a COM-based method to realize the scene segmentation. The experiments prove that the method proposed in this paper is effective.

In this paper, we develop a coarse-to-fine algorithm for recognizing underwater objects. For coarse recognition, a new gray-spatial histogram is proposed, which incorporates spatial information with gray compositions without sacrificing the robustness of traditional gray histograms. For fine recognition, an approach to wavelet transform-based recognition using feature normalization is presented. The combination of coarse and fine recognition reduces the computational cost without degrading the recognizing accuracy. Numerical experiments are given to demonstrate the performance of the new algorithm.

The main characters of classification techniques of multi-spectral remote sensing images are analyzed in this paper. A neural network is constructed and an classification algorithm considering feature texture is proposed. A new method of combining neural network with feature textual knowledge is given to make the algorithm of multi-spectral image classification more effective and practical.

Pesticide is one of the indispensability materials in modern agricultural management, however the excessive use of pesticides has threatened the ecological environment and people's health. This paper introduced an optical instrument to test the pesticide residues in agricultural products based on the inhibition rate of organophosphates against acrtyl-cholinesterase (AchE). The instrument consists mainly of a solid light source with 410nm wavelength, a sampling container, an optical sensor, a temperature sensor, and a MCU based data acquisition board. The light illuminated through the liquid in the sampling container, and the absorptivity was determined by the amount of the pesticide residues in the liquid. This paper involves the design of optical testing system, the data acquisition and calibration of the optical sensor, the design of microcontroller-based electrical board. Tests were done to reveal the affection of temperature and reacting time on AchE, to establish the relationship between the amount of methamidophos and dichlorvos with AchE. The results showed that the absorption rate was related to the pesticide residues and it could be concluded that the pesticide residues exceeded the normal level when the inhibition rate was over 50 percent. The instrument has potential application in vegetable markets.

Offline handwritten chinese character recognition (HCCR) is one of means for quick text input and it has a great demand in the area of file recognition, form processing, machine translation and office automation. However it still is a difficult task for handwritten chinese character recognition to put into practical use because of its large stroke change, writing anomaly, and no stroke ranking information can get, etc. al. An efficient classifier occupies very important position for increasing offline HCCR ratio. Support vector machines offer a theoretically well-founded approach to automated learning of pattern classifiers for mining labeled data sets. But as we know, the performance of SVMs largely depend on the kernel function, different kernel function will produce different SVMs, and may result in different performance. However, there are no theories concerning how to choose good kernel functions based on practical using problem. In this paper we make use of the basic properties of Mercer kernel to construct a hybrid kernel from the existing common kernel, and to find the unknown parameters of the hybrid kernel in data-dependent way by minimizing the upper bound of the VC dimension of the set of functions. Our experiment results show that the proposed method is efficient compared with other classifier for handwritten Chinese character recognition.

CAD-based photogrammetry methodologies have been used in object geometry inspection, and the 3D pose estimation of part is one of the important steps. This paper reports a strategy of 2D-3D pose estimation of sheet metal part by its CAD data and single grey image. It uses Hough strategy to get initial value and then do general point adjustment to get accuracy result. The strategy is tested with the images of some sheet metal parts. It is robust and reliable acting as initial values in certain condition. And it has been used in a sheet metal part inspection system.

In this paper, we use the integrated features of the multispectral image to classify image's texture, namely, the two types parameters are estimated as the texture features: the Hurst parameter and the unit displacement incremental power. The efficiency of the features is evaluated by comparing several other features with them, including the wavelet features. We chose a remote sensing image to make experiment. In this experiment, we see the classification results when the image is partitioned into fewer texture classes and the homogeneity of each class is a little weak. The results show that the effect of the new method in classification is satisfying. It implies that the new features characterize the natural processes successfully and give more texture information.

As a cooperation project between Sweden and China, we sampled a number of rock specimens for analyze rock fracture network by optical image technique. The samples are resin injected, in which way; opened fractures can be seen clearly by means of UV (Ultraviolet) light illumination. In the study period, Recognition of rock fractures is crucial in many rock engineering applications. In order to successfully applying automatic image processing techniques for the problem of automatic (or semi-automatic) rock fracture detection and description, the key (and hardest task) is the automatic detection of fractures robustly in images. When statistical pattern recognition is used to segment a rock joint color image, features of different samples can be learned first, then, each pixel of the image is classified by these features. As the testing result showing, an attribute rock fracture image is segmented satisfactorily by using this way. The method can be widely used for other complicated images too. In this paper, Kernel Fisher discrimination (KFD) is employed to construct a statistical pattern recognition classifier. KFD can transform nonlinear discrimination in an attribute space with high dimension, into linear discrimination in a feature space with low dimension. While one needs not know the detailed mapping form from attribute space to feature space in the process of transformation. It is proved that this method performs well by segmenting complicated rock joint color images.

A new Fuzzy Support Vector Machines (FSVMs) based on linear clustering is proposed in this paper. Its concept comes from the idea of linear clustering, selecting the data points near to the preformed hyperplane, which is formed on the training set including one positive and one negative training samples respectively. The more important samples near to the preformed hyperplane are selected by linear clustering technique, and the new FSVMs are formed on the more important data set. It integrates the merit of two kinds of FSVMs. The membership functions are defined using the relative distance between the data points and the preformed hyperplane during the training process. The fuzzy membership decision functions of multi-class FSVMs adopt the minimal value of all the decision functions of two-class FSVMs. To demonstrate the superiority of our methods, the benchmark data sets of machines learning databases are selected to verify the proposed FSVMs. The experimental results indicate that the proposed FSVMs can reduce the training data and running time, and its recognition rate is greater than or equal to that of FSVMs through selecting a suitable linear clustering parameter.

Template matching has many applications in signal processing, image processing, pattern recognition, and video compressing. It can find a desired template in the large reference image by sliding the template window in a pixel-by-pixel basis, computing the degree of similarity between them, and searching position with the largest similarity measurement. It is computationally expensive to search for every possible position of the template window within the larger reference image. When a rotation exists between the template and the reference image, the conventional template matching algorithm described above is no t practical for real-time processing. In this paper, a fast algorithm is proposed to match the rotated template, which consists of two steps: firstly, the possible rotation angle is detected based on the principle orientation, and then the optimal matching position is searched by matching the edge map. Experimental results on a variety of real images have shown the efficacy of the proposed method.

A fuzzy reinforced learning vector quantization (FRLVQ) algorithm has been recently developed and proved to work successfully in an image compression application. Due to the close affinity between vector quantization and clustering analysis, as both of them group an unlabelled data set into a certain number of clusters such that data within the same cluster have a high degree of similarity, FRLVQ is used as the first stage for clustering analysis in the proposed method, named XHJ-method, in this paper. The simulation results show that this new method works well for the traditional Iris data and an artificial data set based on V29 QAM. Both data sets contain un-equally sized and spaced clusters.

Along with the aero images with high forward overlap and side overlap is becoming increasingly common, a new concept for automated stereo, which named "Multi-Ray-Stereo", has been proposed. Due to the high redundancies of the image data, the multi-image matching makes the fully automatic generation of DSM and true ortho-image become possible. In this paper, a multi-image matching approach for automatic DSM and true ortho-image generation from the aerial images is presented. Several experimental DSMs and true ortho-images are also presented.

The determination of the exterior orientation parameters is an essential pre-requisite in the photogrammetry. In classical photogrammetry this task is solved indirectly by aerial triangulation. The attempt of the reduction or elimination of the ground control points in block adjustment has never been ceased. Because there are abundant parallel lines in the aerial images over the urban areas, according to the theory of vanishing points, the block adjustment combined with the vertical line constraints has been proposed in this paper. The experimental results shows that the block adjustment with vertical lines are more accurate than the block adjustment with control points only, and more accurate than the direct geo-position of GPS/IMU data.

Three dimensional (3D) Surface model reconstruction from 3D scatter points is of great importance in variable fields such as 3D reconstruction based on images, scientific computing visualization, reserve engineering, industry design, analysis and simulation, virtual reality etc.. However, presently 3D points can be obtained with comparatively mature methods such as imagery methods (like photogrammetry, computer vision) and laser scanning etc. while surface model reconstruction from 3D scatter points is still a difficult problem. In this paper, a novel algorithm for automatic generation of 3D surface model based on projective constraint has been proposed. It makes full advantage of projective constraint of points and transfers the 3D Delaunay triangulation problem to a 2D Delaunay triangulation problem. Finally, 3D surface model can be automatically reconstructed. Experiment results show its reliability and robustness.

Camera calibration is the process of determining the intrinsic or internal parameters (i.e. focal length and principal point) of a camera, and is important for both motion estimation and metric reconstruction of 3D models. This paper addresses the problem of calibrating a pinhole camera from images of profile of a revolution. In this paper, the symmetry of images of profiles of revolution has been extensively exploited and a practical and accurate technique of camera calibration from profiles alone has been developed. Compared with traditional techniques for camera calibration which may involve taking images of some precisely machined calibration pattern (such as a calibration grid), or edge detection for determining vanish points which are often far from images center or even don't physically exist, or calculation of fundamental matrix and Kruppa equations which can be numerically unstable, the method presented here used just profiles of revolution, which are commonly found in daily life (e.g. bowls and vases), to make the process easier as a result of the reduced cost and increased accessibility of the calibration objects. This paper firstly analyzed the relationship between the symmetry property of profile of revolution and the intrinsic parameters of a camera, and then showed how to use images of profile of revolution to provide enough information for determining intrinsic parameters. During the process, high-accurate profile extraction algorithm has also been used. Finally, results from real data are presented, demonstrating the efficiency and accuracy of the proposed methods.

The image recorded in a computer vision system depends on three factors: the physical content of the scene, the illumination on the scene, and the characteristics of the camera. The goal of computational color reproduction is to estimate the surface reflectance characteristics, reset the intensities of the color channels in digital images, and render the scene colors canonically. Computational color reproduction has important applications such as object recognition and scene understanding, as well as image reproduction. This paper divides color reproduction into two categories: one is the static color reproduction where the illumination condition is relatively stable; the other is the dynamic color reproduction where the illumination conditions cannot be controlled. We try to summarize and compare the methods respectively for the two categories. The three factors which influence the color values of digital images will be discussed firstly, then typical color reproduction methods for the two categories will be introduced; finally results and conclusion of these methods will be given at the end of this paper.

According to correlative conception and properties of order morphology transformation, non-linear weighted mean value filter is constructed to remove high frequence noise such as noise of Gaussian and impulse. Then an approximated midpoint value filter which can reject uniformly distributed noise is given by improving weighted mean value filter. Based on above, a new image enhancement algorithm is proposed. Within this algorithm, weighted mean value of the image about structuring elements of different directions is calculated and used to identify edge of image; Local average value and entropy difference is applied to control enhancement coefficients. So the target and edge of image are elevated while high frequence noise of image is restrained. The comparison of average value, standard deviation and image entropy between enhanced image and its original illustrates that contrast of image is also improved.

This paper investigate into semi-automatic building recognition in urban area with laser scanned data (LIDAR) and multiple overlap images. Three image sequences, two oblique photography to buildings' walls and one vertical photography to building's roof, and DSM from LIDAR data are used as sources of information. Classification and segmentation can be processed by combined multi-spectral information which is provided by color aerial image and geometric information from a laser scanned DSM . A line segment matching, based on geometry and chromatic constraint, is applied for automatically getting the corresponding line features in multi target images. We also discuss the approach to group 3-D segment into planes and verification of 3D building hypothesis.

In this research, the potencial of using the Visible/Near Infrared Spectroscopy (Vis/NIRS) was investigated for measuring the quality attributes of Tomato 'Heatwave' (Lycopersicum esculentum), and the relationship is established between nondestructive Vis/NIR spectral measurements and the major physiological properties of tomto. The properties include fruit firmness, soluble solids content (SSC) and acidity (pH). Intact tomato fruit were measured by reflectance Vis/NIR in 350- 2500 nm range. The data set as the logarithms of the reflectance reciprocal (absorbance (log1/R)) was analyzed in order to build the best calibration model for each characteristic, using some spectral pretreatments and multivariate calibration techniques such as principal component regression (PCR) and partial least square regression (PLS). The models for the SSC (r= 0.90), standard error of prediction (SEP)= 0.19 °Brix with three factors; pH (r= 0.83), standard error of prediction (SEP)= 0.09°Brix with factors, compression force r= 0.81), standard error of prediction (SEP) =16.017°Brix N with six factors, and puncture force (r= 0.83), standard error of prediction (SEP) = 1.18°Brix N with three factors, shown the good to excellent prediction performance. The Vis/NIR spectroscopy technique had significantly greater accuracy for determining the SSC. It was concluded that the Vis/NIRS measurement technique seems reliable to assess the quality attributes of tomato nondestructively.

The symbology of the PDF417 barcode was introduced. Based on the difference in the direction to the vanishing point of the columns boundaries, an algorithm for detecting the rows boundary of the PDF417 barcode was proposed. This algorithm can detect the columns boundary and the rows boundary of the PDF417 barcode and locate the codewords position exactly, even though the minimum bar/space width of the PDF417 barcode is only about 1.5 times the pixel width or the barcode image is demolished severely.

To overcome restriction of close modeling techniques, and to approach a feasible close-range photogrammetry modeling, we present a realistic system on the basis of multi-baseline stereo vision. The system incorporates the image matching based on short-baseline-multi-views, and 3D measurement based on multi-ray intersection, and the 3D reconstruction of the object's integral model. One plane is reconstructed by this approach, the results demonstrates the feasibility and effectivity.

In this paper, automatic reconstruction of building from DSM is presented. Focusing of ROI (regions of interest) is a pivotal step in reconstruction. Contour tracing is adopted here to get the ROI and every layer of building's closed boundary is acquired simultaneously. The normal vector analysis of ROI is used to select the geometric model and reconstruct the roof. The closed boundaries are adjusted by least square refining to constitute the buildings of multi right angles and multi layers. The geometric models used are prismatic (flat roof) and parametric (gabled roof). The experiment shows that the result of reconstruction is satisfactory.

In this paper, an algorithm of automatic mosaic of facade texture from vehicle-based close-ranged image sequence is presented. Close-ranged DC images have nonlinear distortion due to the large obliquity of the posture of DC and obvious lens distortion, so it needs to be implemented that the preprocessing of automatic rectification of images, Occlusion removal and elimination of lens distortion. As close-ranged imagery as concerned, the facade is too large to be covered by a single image. Therefore, image mosaic is indispensable to acquire whole facade texture. Two mosaic methods are presented in this paper. 1) Selecting mosaic point pair. After the preprocessing, only the displacements in X-axis and Y-axis respectively are needed to determine in mosaic process because facade textures are all rectified onto plumb plane. Therefore the strategy combined with correlation coefficient and geometric constraint is employed to acquire mosaic point pair by which the displacements are determined. This method cannot be applicable when the imagery is taken too close to the facade (< 50m), because the geometric distortion cannot be eliminated perfectly after the rectification process. 2) Strip method. It is deduced to determine the relatively spatial relationship of the images facade texture involved that a model of least square adjustment, similar to the strip block adjustment, based on the formula of the projection of image point on the projective plane with the DC angular parameters determined early as weighted observation. Then a method similar to generating ortho-image is employed to generate the whole facade texture. This method is proved to be more applicable.

An effective feature fusion strategy applied in vehicles classification is devised, which takes advantage of the complementary vehicle features in Synthetic aperture radar (SAR) and Optical images. With high spatial resolution SAR images, it is easy to detect vehicles fast and accurately because of the strong radar cross sections (RCS) of them compared to background. However, the classification of vehicles in SAR images can carry a significant amount of error (misclassification) since the radar scattering from a vehicle is often highly dependent on the target-sensor orientation. In contrast, optical images usually provide good classification performance exploiting the aspect dependent information. Therefore, the proposed method maps the detection results of SAR image into the co-registered optical image, and then combines the target's features from SAR and Optical images together to feature-fusion classification using the Fuzzy C-means (FCM) algorithm.

At present neural network models make great progress in remote sensing image classification, but these models have some serious limitations, such as easily getting stuck at a local minimum, converging too slowly and uneasy-fixed network structure. Support vector machines (SVMs) is a nonlinear mapping algorithm based on the Statistical Learning Theory, and developed over the last three decades by Vapnik, Chervonenkis et al. It gained extensive applications in pattern recognition and regression analysis etc. Compared the SVM algorithm with neural network models, the former is based on self-contained mathematics theory, and furthermore solves a global optimization problem and makes sure the result is not local minimum. These enable the SVM algorithm excellent classification performance. The paper proposed a new hybrid classification method that combines support vector machines with fuzzy membership function for remote sensing image. Firstly the method constructs multi-class Support Vector Machines classifier for remote sensing image, and discusses parameter estimation problem, and then uses RBF kernel SVM to classify whole remote sensing image. Secondly aiming at the disadvantage of SVM classifier that exists some mixed samples (one sample divided into two or more categories) and missed samples (one sample is not classified), and using fuzzy membership function method to reclassify these mixed and missed samples. Experimental results suggested the accuracy of this hybrid classifier is higher than single SVM method, or single fuzzy membership function decision method or BP neural network model.

This paper presents the analogy between voice recognition and multi-spectral remote sensing image classification, and introduces the Hidden Markov Model (HMM), which is a successful approach on voice recognition fields, into multi-spectral remote sensing image classification. After comparing the HMM with other conventional classification methods such as Maximum Likelihood and Minimum Distance, the paper concludes that the HMM is a better approach than other techniques do. At the end of the paper, the author explains the reason of HMM's good performance, and also points out its defect.

Kansei image retrieval is a new kind of retrieval technology with high complexity. However, it's likely that only some parts of the image would attract people and produce affections. Color imposes a great impact upon the feeling as the basic feature of image, and the entropy of the image also exhibits the information quantity. In this paper, we present a method of kansei image retrieval utilizing the color and entropy to extract regions of interest (ROI). Back propagation neural network is employed to map the color and entropy of ROI to affective feature space. Finally, we show some experimental results of ROI extraction and kansei image retrieval based on interest.

Radial basis function(RBF) neural network was applied to determine soil types of hilly and mountainous terrains in Fengdu County of the Three Gorges region in China, the elevation in which ranges between 118.5m and 2000m, combining landsat enhanced thematic mapper plus (ETM+) data and topographic information from a digital elevation model (DEM). We designed a RBF network using newrb P,T,GOAL,SPREAD) function in MATLAB software, in newrb function orthogonal least squares learning algorithm be used to choose Gaussian kernel function centers and the weights of the network. Two sets of training samples were selected for training. One was a set of 3606 training samples; the other was a set of 57905 training samples, also for maximum likelihood classification. Considering training time, we divided these 57905 samples into 3063 small sample areas, so a set of averages of which was selected to input the network for training at last. The classification results with RBF neural network showed that the second training samples set generated 60.3% producer's accuracy, higher than that of the first samples set. But the producer's accuracies of RBF neural network trained by both sets were lower than that of using maximum likelihood classifier with the same training samples, which was 66.6%. On the other hand, the Kappa coefficient of RBF neural network trained by the second training samples set was 0.5587, higher than that of maximum likelihood classifier with the same training samples, which was 0.4919. So, it is indicated that RBF neural network for soil classification is not the best method under limited training samples for more samples more training time consumption to the unacceptable extent.

In this paper, a Kansei landscape image retrieval system named KIRCK is proposed, which is based on color feature and Kansei factors. Color feature is extracted in HSV color space and the similarity of color feature is estimated by color accumulation histogram intersection method. Multi-class Support Vector Machine is applied for the mapping between high-level Kansei labels and low-level image characteristics. After the multi-class SVM is trained, Kansei factors of images can be labeled automatically, and the similarity of images in Kansei space also can be estimated. Thus integrated retrieval results using color and Kansei factors can be obtained, and the experiment shows that these retrieval results are more satisfied than only using color feature or Kansei factors. Correlative feedback is also introduced to improve the performance of our color feature and Kansei factors image retrieval.

The digital camera and the slide projector are being frequently used in broad range of the photogrammetric measurements with the development of non-contact measurement in the close-range photogrammetry. In order to take full advantages of the slide projector, the slide projector is steered on the basis of the traditional method of the digital camera taking images. A slide projector and a digital camera form the projector-camera system, which is similar with a binocular vision system in the principle of 3D reconstruction. Before using the projector-camera system, the interior parameters of both have to be calibrated in advance. The paper proposes a flexible and effective technique for the united calibration of the digital camera and the slide projector. The technique only requires a planar grid, a rotating platform and a computer. The algorithm with 2D direct linear transformation (2D-DLT) and collinear equations is used to calibrate the digital camera first. By the similar algorithm above, the interior parameters of the slide projector can be determined too. Then the united calibration of the digital camera and the slide projector is achieved finally. The operation method in detail and the algorithm are addressed systematically and entirely. The feasibility and the exactness of the technique for the united calibration of the digital camera and the slide projector put forward in this paper are verified by the results of real data.

When an monocular vision-based unmanned aerial vehicle (UAV) based on vision is flown to the final approach fix to intercept the glide slope without the navigation of Global Positioning System (GPS), the position and orientation of the airport runway in image must be detected accurately so as to a host of suitable procedures have to be followed. The optimum length of the final approach is about five miles from the runway threshold. The front view of the runway, which is achieved at the moment, is very illegible. The approaching marking (cross bar) of the runway are showed as some white spots of high intensity and the complicated backgrounds of the airport are included in the images. In this case, spots with high intensity should be extracted and classified, some of these spots are just the images of the background noises and the pseudo-targets, which can't be separated with the spots of the runway as in the view there is no significant characteristic difference among them ostensibly. Fortunately, in the terrestrial coordinate space, most of the runway marks are located at the apexes of a rectangle, having some geometric relationships. The relationship among the projection coordinates of the runway spots in the images can be determined according to the perspective principle, the constraint condition of the rectangle as well as the front shot constraint condition of the target, by using this relationship, the runway approaching marks can be separated, the position and the direction of the runway in the images can be identified. In this paper, the clustering management is adopted so as to greatly reduce the computing time. The consequence of the experiments shows that by this algorithm, even from a place far away from the runway whose marks are unclear, we also can effectively detect the runway.

As computer graphics technology becomes more popular, a large amount of 3-D model data is required for various purposes. Three-dimensional (3-D) digital modeling of object through 3-D optical scanners has been demonstrated in recent years. However, most current 3-D scanners are usually large, heavy, expensive, and commonly accurate (Mean Square Error is around 0.3mm). Presently, highly accurate, easy-to-use and cheap 3-D shape acquisition system provides viable solutions to meet the requirements of users. In this paper, we introduce a linear structured scanning system covers an area of 600mm by 400mm in width in a high accurate way (MSE is lower than 0.1mm). Hardware setup of the linear structured scanning system and the process of the shape acquisition for 3-D object modeling based on the linear structured scanning system are proposed in this paper. We also proposed a highly accurate calibration method which can overcome complicated calculations of camera parameters. This highly accurate calibration method can be widely applied to linear structured light measurement system. Then we describe the modeling approach of our linear structured scanning system. Finally, reconstruction and visualization results are illustrated and further analyses are discussed in this paper.

Recently, the application of artificial neural networks (ANN) into classification has raised a great deal of interest. The standard back-propagation (BP) algorithm is suitable for training neural networks. Generally, the topological structure of the back-propagation neural network (BPNN) used in the image classification consists of neurons arranged into three layers, namely input layer, hidden layer and output layer. Obviously, the number of input layer nodes in a BPNN generally corresponds to the number of features (spectral bands), which influences the iterative (or training) time. Nowadays, sensors provide more and more spectral bands. Hence, the representation of multi-spectral remote sensing data in ANN has become a mayor problem. Selection of the effective image bands in order to reduce the size of the input data is therefore necessary using, for example, the Principal Component Analysis (PCA). In this paper, an improved BPNN model, using the PCA and BPNN, has been developed. This proposed method can reduce the number of input layer nodes aiming at attaining the effective "bands" for classification. The experimental results show that the proposed method can cut down computational costs during training stage because of the reduction of the number of input nodes and improve the overall classification accuracy.

Fingerprint verification is one of the most reliable personal identification methods. However, most of the automatic fingerprint identification system (AFIS) is not run via Internet/Intranet environment to meet today's increasing Electric commerce requirements. This paper describes the design and implementation of the archetype system of identity authentication based on fingerprint biometrics technology, and the system can run via Internet environment. And in our system the COM and ASP technology are used to integrate Fingerprint technology with Web database technology, The Fingerprint image preprocessing algorithms are programmed into COM, which deployed on the internet information server. The system's design and structure are proposed, and the key points are discussed. The prototype system of identity authentication based on Fingerprint have been successfully tested and evaluated on our university's distant education applications in an internet environment.

Document classification is one of important steps in document mining. In this paper, we present a new kind of document classification method based on generalized learning model (GLM for short). GLM is an extensible machine learning model with great flexibility. It may fuses symbolic learning, fuzzy learning, statistical learning, and neural learning together. If necessary, new learning model can be incorporated. To describe and represent documents more reasonably, we develop a approach to extract membership vector as features of documents. In view of the characteristics of document classification, two kinds of document classification methods are employed under GLM frame. One is based on fuzzy set theory, the other is based on support vector machine (SVM). These two kinds of methods can supplement each other to achieve better performance.

In this paper we present an approach to extract semantic object based on color feature using ISODATA clustering algorithm. First, we translate RGB color space into L*a*b* color space. Second, we use the ISODATA algorithm to solve clustering problem. In the end, we extract semantic object in terms of color information. Experimental results show that the color clustering give superior results in increases in cluster compactness.

Bayesian Networks have emerged in recent years as a powerful data mining technique for handling uncertainty in complex domains. The Bayesian Network represents the joint probability distribution and domain (or expert) knowledge in a compact way and provides a comprehensive method of representing relationships and influences among nodes (variables) with a graphical diagram. Actually, however, in the classification domain it was not paid attention to by researchers until the simplest of form of Bayesian Networks, Naive Bayes Classifier, turned up. Naive Bayes Classifier is a simple and efficient probability classification method, and has shown surprising performance in some domains, which owes to the independence assumption that makes Naive Bayes Classifier fit the classification more easily. However, the independence assumption obviously does not hold in the real world. Therefore, in order to meet the "naive" (unreal) assumption, this paper proposes a new image texture classification method of aerial images, PCA-NBC, which combines the Principal Components Analysis (PCA) and Naive Bayes Classifier (NBC). The PCA transforms the highly correlated features into statistically independent and orthogonal "features", so it is suitable to solve that problem and can lay a solid theoretic foundation in the application. One hundred and thirteen aerial images are used to evaluate the classification performance in the experiment. The experimental results demonstrate that the proposed method can cut down the number of features and computational costs and improve the accuracy during classification. In one word, the new method, PCA-NBC, is an attractive and effective method, which outperforms the Naive Bayes Classifier.

This paper points out, by data analysis, that high model number and poor elevation accuracy of mapping are two problems of the digital aerial camera, and presents methods of map-based mapping and multi-baseline stereo matching to solve the problems. Finally an improved photogrammetric workflow and some difficult technology problems are presented.

Aiming at the problem that the existing methods have difficulty to identify extracted watermark after signal process, this paper proposes a novel method based on synergetic pattern recognition. Synergetics offers an approach, from top to down, to the constructions of highly parallel scheme for pattern recognition and decision making. According to the synopsis of the usual watermark system, an effective recognition method has been implemented based on synergetic approach. The algorithm model is discussed in detail and used in the extracted image watermark identification. With lots of samples in the spatial and DCT domain watermark, experimental results show that the proposed method is more effective, fast and robust.

The artificial neutral network is a type of large-scale nonlinear parallelism system, capable of identifying causal relationships between complex variables. This paper presented the relationships between the winter wheat yield and the soil spatial distribution information, including water content, organic matter, total nitrogen, alkali-hydrolysable nitrogen, rapidly available phosphorus and potassium, by training 50 tested soil samples in the back-propagation neutral network of topological structure 6:9:1. After verifying the model by the remaining 13 samples, the results show that the soil water content and alkali-hydrolysable nitrogen are linear to the crop yield, the total nitrogen, organic matter and rapidly available potassium are respectively multinomial to it and that the rapidly available phosphorous is of the exponential relationship with the crop yield.

Because wear particles have fractal characteristics, it is necessary that adding fractal parameters to studying wear particles and diagnosing machine troubles. This paper discusses fractal parameters of wear particles, presents arithmetic calculating fractal dimension, and constructs a fractal neural network which can recognize wear particles image. It is proved by experiments that this fractal neural network can recognize some characteristics of wear particles image, and can also classify wear types.

As System-on-Chip (SOC) manufacture technology moves into ultra deep sub-micron (DSM) ear, Crosstalk faults between SoC interconnect result in improper function of the chip. This problem is becoming more and more severe. Based on the in-depth research of the property of crosstalk fault and the MAF model, we presented a simple and efficient model: the Search-Based Maximal Aggressor Fault (SB-MAF) for detecting glitch and delay faults caused by crosstalk effects on interconnects between components of a SOC. The respective efficiency of the presented model and the MAF model is given in the paper. The results of simulation show that two models' efficiency is comparable when crosstalk is weak. However, the efficiency of the SB-MAF model is obviously improved compare to the MAF model when there are strong crosstalk effects between SoC interconnects.

On the basis of the rigorous theories of the digital photogrammetry and the latest research achievements in computer vision and image processing pattern recognition, it is studied for three dimensional(3-D) reconstruction of irregular industrial sheetmetal parts by using the projector and CCD camera in this paper. As the surface of sheetmetal parts is short of the applicable texture for image matching, it is designed that structured light from the projector is projected onto the surface of sheetmetal parts as it were put on an available texture artificially. Supposed that the image acquired from CCD camera is the left image and the suppositional image from the projector is the right image. Then the image acquired from CCD camera is made edge detection, in which the extracted curves are made curve matching with the suppositional lines from the projector. While made fully use of the feature of the right image obtained from the projector and collinearity equations, the spacial point coordinates on the surface of the sheetmetal parts are obtained quickly by space intersection. Then 3-D reconstruction of the sheetmetal parts is completed after merging all the models. The untouched method which has high efficiency and strong flexibility can be successfully adapted to reconstruct various kinds of shapes and some soft objects.

The specific establishment process of a quasi-2D control field and a transformation and its mathematic model from the quasi-2D control field to a virtual real-2D field are introduced in this paper. The influence on the upper coordinate transformation, which is caused by the surveying error of camera position, is analyzed. Particularly, a means of creating the digital distortion model (DDM) and the principle of using the model to correct the image point deformations with various errors totally are put forward. The method, which set up the quasi-2D control field, is simple, convenient and robust. The maintenance of the quasi-2D control field is also easy.

The granular appearance of speckle noise in synthetic aperture radar (SAR) imagery makes it very difficult to visually and automatically interpret SAR data. Therefore, speckle reduction is a prerequisite for many SAR image processing tasks. We develop a speckle reduction algorithm by fusing the wavelet denoising technique with support vector machine (SVM). Based on the least squares support vector machine (LS-SVM) with Gaussian radial basis function kernel, a new denoising operators used in the wavelet domain are obtained. Simulated SAR images and real SAR images are used to evaluate the denoising performance of our proposed algorithm along with another wavelet-based denoising algorithm, as well as the refined Lee speckle filter. Experimental results show that the that the proposed filter method outperforms standard wavelet denoising techniques in terms of the ratio images and the equivalent-number-of-looks measures in most cases. It also achieves better performance than the refined Lee filter.

In this paper, we propose an unsupervised change-detection method which considers the spatial contextual information in a log-ratio difference image generated from multitemporal SAR images. A Markov Random Filed (MRF) model is particularly employed to exploit statistical spatial correlation of intensity levels among neighboring pixels. Under the assumption of independency of pixels each other and mixed Gaussian distribution in the log-ratio difference image, a stochastic and iterative EM-MPM change-detection algorithm based on a MRFs model is developed. The EM-MPM algorithm is based on a maximiser of posterior marginals (MPM) algorithm for image segmentation and an Expectation-maximum (EM) algorithm for parameter estimation in completely automatic way. The experiment results obtained on multitemporal ERS-2 SAR images show the effectiveness of the proposed method.

Forest biomass is one of the most important parameters for global carbon stock model yet can only be estimated with great uncertainties. Remote sensing, especially SAR data can offers the possibility of providing relatively accurate forest biomass estimations at a lower cost than inventory in study tropical forest. The goal of this research was to compare the sensitivity of forest biomass to Landsat TM and RADARSAT-SAR data and to assess the efficiency of NDVI, EVI and other vegetation indices in study forest biomass based on the field survey date and GIS in south china. Based on vegetation indices and factor analysis, multiple regression and neural networks were developed for biomass estimation for each species of the plantation. For each species, the better relationships between the biomass predicted and that measured from field survey was obtained with a neural network developed for the species. The relationship between predicted and measured biomass derived from vegetation indices differed between species. This study concludes that single band and many vegetation indices are weakly correlated with selected forest biomass. RADARSAT-SAR Backscatter coefficient has a relatively good logarithmic correlation with forest biomass, but neither TM spectral bands nor vegetation indices alone are sufficient to establish an efficient model for biomass estimation due to the saturation of bands and vegetation indices, multiple regression models that consist of spectral and environment variables improve biomass estimation performance. Comparing with TM, a relatively well estimation result can be achieved by RADARSAT-SAR, but all had limitations in tropical forest biomass estimation. The estimation results obtained are not accurate enough for forest management purposes at the forest stand level. However, the approximate volume estimates derived by the method can be useful in areas where no other forest information is available. Therefore, this paper provides a better understanding of relationships of remote sensing data and forest stand parameters used in forest parameter estimation models.

A new evaluation method is proposed for single polarimetric speckle filters of Synthetic Aperture Radar (SAR) images, including two steps: measurement and aggregating multicriteria. In accordance with three properties of a good speckle filter: speckle reduction, feature preservation and radiometric preservation, six measures are selected: Equivalent Number of Looks (ENL), Target-to clutter Ratio, Radiometric Loss, Bias of Mean Value, Bias of Spatial Resolution, Bias of Peak Sidelobe Ratio, where the last two criteria are firstly addressed in this presentation. In multicriteria decision making, Ordered Weighted Averaging (OWA) operator is used to aggregate the performance preference matrix induced by measures. The experiment result of real and synthetic SAR images for five speckle filters conforms to that of visual comparison.

Regional surface subsidence induced by underground coal mining is very common and is a serious environmental problem in China. The mining subsidence not only causes damages to surface buildings but also change the pattern of surface drainage in a densely urbanized area. Monitoring and Analyzing the spatial distribution of the endangered surface may be helpful for land-use planning and for land reclamation. Interferometry SAR (InSAR) can be used to effectively monitor the succession of the spatial extent and the magnitude of subsidence in coal mining areas. In this paper, the multi-temporal D-InSAR method was applied for the generation of deformation map in coal mining area. With the "Interferometry Coherence Estimation Minimum Span Tree (ICEMST)" model, the optimized SAR images combination for D-InSAR processing for long term surface subsidence monitoring were predicted. With the estimation of ICEMST, several scenes SAR SLC data with time spanned more than half a year and spatial baseline more than 400m long were combined for D-InSAR processing to study the succession of land subsidence induced by underground coal mining and groundwater exploration in the test site, Kailuan coal mine area, a typical mining industrial area in north China, which has 125 years coal mining history. After being processed with the conventional "2 Pass" differential InSAR method, the deformation caused by underground coal mining in the line of sight (LOS) was transformed into vertical subsidence map. The experiment shows that the short time span is more suitable for D-InSAR application in mining areas than the long time span, for the lower correlation due to densely growing vegetation, seasonal changing factors and large water plashes. The time decorrelation, spatial decorrelation and the D-InSAR error resources were analyzed and discussed, and the Connor Reflectors method integrated with D-InSAR and GPS are presented, which will be a key practical technology for information obtain in digital mine.

Yellow River is the second longest river in China, but from of old, its annual ice flood threatens people's safety and property in the Yellow River basin. Yellow River's ice flood is due to its especial geographic condition with its flowing direction from low latitude to high latitude. Every year the government spends much of manpower and material resource in Yellow River ice flood prevention. It was the first time for Yellow River Conservancy Commission (YRCC) to use Remote Sensing data in the Yellow River daily ice detection, and MODIS data as the primary data in this monitoring work. For the high temporal resolution, MODIS (the Moderate Resolution Imaging Spectroradiometer) data is useful in monitoring ice flood changing day by day. The MODIS snow and ice cover algorithm is based on the reflectance characteristic of snow. In this paper, two criteria, the Normalized Difference Snow Index (NDSI) and one test of near-infrared band's reflectance were used to identify Yellow River's ice. By comparing to the ground truth information from Yellow River hydrologic stations, the algorithm was developed, the result validated, the achievements described, and the shortage analyzed. The results show that MODIS is efficient in ice and snow detection.

A registration method combining watershed segmentation and mutual information for Synthetic Aperture Radar(SAR) images is introduced in this paper. In this approach, the gradient images of SAR are first segmented by morphological watershed algorithm. Oversegmentation is diminished by basin dynamics. Mutual information is calculated on a pair of segmentation results. The maximum of the mutual information indicates the optimal matching parameters. In order to avoid falling into local optimal, a two-step modification is also proposed. Coarse parameters are obtained on the segmentation result containing a small number of regions, and refinement is done at the neighbour of the coarse matching with more regions in segmentation. The optimal matching at the second step indicates the final registration parameters. Experiments on multi-polarization SAR images show that this approach has evident effect for registration of SAR images with rigid or small geometrical difference.

A segmentation method using maximization of the Posterior marginals (MPM) and constrained stochastic relaxation (CSR) for SAR images is proposed. This method improves the regularity of MPM based segmentation result by introducing CSR. Multi-Level Logistic (MLL) model is used for the underlying label image to introduce regularity prior of segmentation. Gamma distribution is used for SAR intensity data. The hyper parameters of MLL model are supposed to be known a priori. This method is an iterative scheme consists of two alternating steps: to approximate the MPM estimation of the pixel class labels and to estimate gamma distribution parameters. The weight of the prior energy in goal energy function is increased slowly versus the increasing iteration times until certain number of iteration has finished. The segmentation results for synthetic and real SAR images show that the proposed method has a good performance.

Uncompensated phase errors presented in synthetic-aperture-radar (SAR) data have a disastrous effect on SAR image quality. To estimate and compensate phase errors, a new method is presented based on the time-frequency distributions of the range-compressed SAR signal. Robust phase errors estimates are obtained by utilizing range redundancies The processing results of the simulated data show the validity of the proposed method.

A pixel-level image fusion method based on turbo iterative for Synthetic Aperture Radar (SAR) and other sensor images is proposed. In this method, a wavelet-based fusion algorithm is employed at first. Then the output image is used to estimate the prior information of the ideal fusion result which is needed in the next step of adoption of a model-based fusion method. Such iterative is performed until the export image's change is at the convergence. The fusion results for SAR images show that the proposed method has a good performance.

There are many cities in China suffering seriously from land subsidence caused by groundwater over-extraction, but sufficient money has been lacking in monitoring the subsidence. Compared with leveling and GPS surveying, D-InSAR is more precise and cost-efficient. Tianjin, a city with smooth topography and severe subsidence, is an ideal place for testing and analyzing D-InSAR technique, especially with plenty of Envisat and ERS-1/2 SAR data and thousands of valuable leveling data being available. However, D-InSAR is liable to be contaminated by atmosphere delay, temporal decorrelation and baseline errors. These errors cannot be removed by SAR data processing. This paper analyzes the features of atmosphere delay with the auxiliary data in Tianjin urban area. The results demonstrate that D-InSAR can detect subsidence within three months. Further actions have been undertaken to improve the D-InSAR subsidence monitoring system, such as using Envisat data, gathering GPS zenith delay data, etc. to reduce errors. The final interest of our research is to establish a robust, cost efficient city subsidence monitoring system by using D-InSAR technique.

The inverse synthetic aperture radar (ISAR) imaging of non-uniformly rotating target and maneuvering target is a complicated task and the conventional range-Doppler (RD) ISAR technique does not work properly in this case. A novel ISAR technique, named range-instantaneous Doppler (RID), has been proposed to deal with this difficulty. In this paper, the fractional Fourier transform (FrFT) and the CLEAN technique are combined for RID imaging and the detailed algorithm is presented. We use computer simulation to demonstrate its performance and the simulation results show its effectiveness.

Registration of two or more images of the same scene is an important procedure in INSAR image processing that seeks to extract differential phase information not obtainable from each one of these images. Meanwhile, the accuracy of this step is crucial to the reliability of subsequent image processing and final results of the data processing chain. Based on some conventional INSAR registration methods, this paper presents an approach integrating correlation-registration and least square-registration to attain sub-pixel precision. Furthermore, experimentations implemented on test site prove validity of the registration method. Finally, some significant conclusions are made by the experiment results.

In this paper, an improved adaptive speckle reduction method is presented based on dual tree complex wavelet transform (CWT). It combines the characteristics of additive noise reduction of soft thresholding with the CWT's directional selectivity, being its main contribution to adapt the effective threshold to preserve the edge detail. A Bayesian estimator is applied to the decomposed data also to estimate the best value for the noise-free complex wavelet coefficients. This estimation is based on alpha-stable and Gaussian distribution hypotheses for complex wavelet coefficients of the signal and noise, respectively. Experimental results show that the denoising performance is among the state-of-the-art techniques based on real discrete wavelet transform (DWT).

For detection of moving targets with single-antenna SAR, a number of algorithms have been proposed. But the detection of slowly moving target in the strong ground clutter is still a problem. In this paper, a new method has been presented for resolving above problem. The ground clutter has been primarily cancelled based on multi-look processing technique. The slowly moving targets are isolated from the strong ground clutter at last by using the SD Intensity Contrast Function.

Basis Pursuit is a novel technique of signal sparse representation. It seeks sparse representation from over-complete dictionaries. In this paper, we extend the application of the Basis Pursuit method to SAR super-resolution imaging. Firstly, based on the SAR attributed scattering model and adaptive fine grid, a new type compacted Dictionary is designed in phase history domain. By this way, the dimension of the desired problem is even smaller. Secondly, a new and fast iterative algorithm is proposed, the fine feature parameter estimation of the scatter in SAR image is obtained. Finally, in terms of SAR attributed scattering model in the phase history domain, larger scale phase history data is built. By FFT imaging, higher resolution image is obtained. Simulation experiments and computational results of measured MSTAR data demonstrate that Basis Pursuit can be performed speedy and stably, it can provide super-resolution at the same time.

Traditional filtering methods can't work well in Synthetic Aperture Radar (SAR) image speckle reduction due to the existence of the speckle noise. In the paper, a new method to reduce SAR image speckle is proposed by analyzing the distribution characteristic of SAR image in wavelet domain and existing spatial filtering methods. Firstly, logarithmically transformed SAR image is decomposed with wavelet transform. Secondly, to overcome the difficulty of choosing the threshold in SAR image speckle reduction based on wavelet transform, the child images decomposed are filtered with fuzzy median value filter whose filter parameters can be self-selected. Finally, according to the child images filtered, we reconstruct the image and give exponential transform, and get the SAR image filtered. The experiment results show that the new method is efficient in SAR speckle reduction.

In this paper we present a new algorithm of road object recognition in SAR image. This algorithm fully makes use of the properties of SAR images. It employs an adaptive mean filter to depress the coherent speckle noise and preserves the edge in the low-level processing, which makes the road extraction much easier. In the middle-level processing, we designed an oriented filter to extract the potential road objects according to the continuity of roads in gray and direction, and in the end, road recognition is utilized in the high-level processing. The algorithm we present is verified to be effective when it's applied to the road extraction using the Radarsat Image of Taiwan district, China.

The gray scale data in SAR image samples containing vehicle targets and common image quantization and enhancement methods have been generally analyzed and a new SAR image quantization and enhancement method in the spatial domain based on point processing has been given, in which, image data quantization range is selected by using the statistical features in gray data of all image sets as parameters and gray scales of the probable target points with high scatting strength beyond the quantization range are set to the highest gray scale. Thereby not only the original statistics in images are retained but also the target information is enhanced. The method of optimal parameter selection is also given. The well-pleasing results for vehicle target recognition have been obtained.

Synthetic aperture radar (SAR) can produce high-resolution images if illuminated terrain by coherently processing the data collected by a single antenna at different locations. In SAR imaging, Scatterers often perform the frequency and aspect angle dependent reflectivity characteristic, which is usually neglected in the traditional imaging processing. The target detection and recognition process requires additional processing to form enhanced SAR images, in which features of man-made objects are easier to identify. In this paper, a novel time-frequency representation imaging formation (TFRIF) is used to extract the frequency and aspect-dependent information without loss of azimuth resolution, which gives additional features in the following targets detection and recognition processing. Simulation results validate the efficiency of the TFR method.

This paper presents a method to produce DEM of steep mountains from InSAR interferogram without using any GCP. The SRTM data of the same area are used to assist the conversion from unwrapped phase to height. It has been proved that the main error sources of InSAR DEM generation are interferential phase and baseline estimation. The systemic part of these errors can be removed by some ground control points (GCPs). However in wilderness it's too difficult to build and identify any GCP. Since 3 arc seconds resolution SRTM data of the most part of the world have been in public by now, instead of GCPs, we improve the quality of InSAR DEM in mountainous area by using SRTM data. In this method, certain points with higher coherence and lower phase slope are selected first from the unwrapped interferogram. An iterative process is used on these points for calculating the phase difference in slant range space between InSAR and SRTM data. The points converged in certain stable location are selected finally. Through these points the phase trends is removed by linear regression and the baseline in the whole image is modeled. Then, the different height between InSAR and SRTM data are calculated pixel by pixel, and these height differences are added to SRTM data to get each InSAR pixel's height, latitude and longitude in WGS84. Experiment result using ERS-1/2 data, which is compared with authoritative 1:50000 DEM over Three Gorges, China, is given.