By suitably phase-encoding optical images in the pupil plane and then digitally restoring them, one can greatly improve their quality. The use of a cubic phase mask originated by Dowski and Cathey to enhance the depth of focus in the images of 3-d scenes is a classic example of this powerful approach. By using the Strehl ratio as a measure of image quality, we propose tailoring the pupil phase profile by minimizing the sensitivity of the quality of the phase-encoded image of a point source to both its lateral and longitudinal coordinates. Our approach ensures that the encoded image will be formed under a nearly shift-invariant imaging condition, which can then be digitally restored to a high overall quality nearly free from the aberrations and limited depth of focus of a traditional imaging system. We also introduce an alternative measure of sensitivity that is based on the concept of Fisher information. In order to demonstrate the validity of our general approach, we present results of computer simulations that include the limitations imposed by detector noise.

We extend a recently introduced universal grayscale image quality index to the perceptually decorrelated color space. The resulting color image quality index quantifies the distortion of a processed color image relative to its original version. We evaluated the new color image quality metric through observer experiments in which subjects ranked images according to perceived distortion. The metric correlates strongly with human perception and can therefore be used to assess the performance of color image coding and compression schemes, color image enhancement algorithms, synthetic color image generators, and color image fusion schemes.

The advancement of non-linear processing methods for generic automatic clarification of turbid imagery has led us from extensions of entirely passive multiscale Retinex processing to a new framework of active measurement and control of the enhancement process called the Visual Servo. In the process of testing this new non-linear computational scheme, we have identified that feature visibility limits in the post-enhancement image now simplify to a single signal-to-noise figure of merit: a feature is visible if the feature-background signal difference is greater than the RMS noise level. In other words, a signal-to-noise limit of approximately unity constitutes a lower limit on feature visibility.

Commercial high-resolution satellite imaging sensors are used for precision mapping, analysis and earth science applications. The evaluation of their imaging performance on-orbit requires special ground targets. Urban areas are rich with potential image quality targets with simple geometries, scuh as points, lines and edges. In this paper, we use parking lot stripes to evaluate the Line Spread Function (LSF) of the DigitalGlobe QuickBird sensor and compare the results to those obtained from edge targets. The two techniques produce equivalent LSF measurements. Parking lot stripes are therefore an alternative target that have some advantages and disadvantages compared to edges.

When a linear image acquisition device captures an image, several noise sources impact the quality of the final image that can be digitally procesed. We have previously examined the impact of these noise sources in terms of their impact on the total amount of information that is contained in the image, and in terms of their impact on the restorability of the image data. In this paper, we will examine the effect of each of the noise sources on final image quality.

Images associated with Underwater Imaging Systems are frequently degraded due to absorption and scattering effects from its underwater environment. The absorption effect reduces the signal strength, and the latter effect reduces both signal strength and image resolution. The optimization of underwater imaging system parameters predominantly focuses on maximizing signal strength and minimizes scattering effects. In the domain of underwater images, the assessment of image quality is highly subjective and lacks the availability of a "standard" objective criteria, which allows a comparison of different imaging techniques and their effectiveness to be performed in a more objective manner. This paper focuses on an experimental performance evaluation of underwater imaging system through objective image quality measurement. The technique is based on 2 dimensional grayscale image of USAF (United State Air Force) target. These targets have been used extensively in underwater imaging system development. It has resolution bars in various frequencies and arrangement, which enable spatial frequencies and signal strength analysis. 3 different image evaluation techniques are used to quantify image quality of an Underwater Imaging System in increased turbidity condition. Peak Signal to Noise Ratio (PSNR) measures the output signal over its Mean Square Error (MSE). Fidelity, F, represents the absolute difference of amplitude values and the Correlation coefficient, r, reflects only changes of signal shape. The metrics scales are considered as a measure of similarity between the idealized and the derived image. In these techniques, the indices are applied to corresponding pixels of the derived and ideal images. The fidelity, F and correlation coefficient, r are applied to thermal imaging evaluation by Volodymyr Borovytsky and Valery Fesenko in 2000. The underwater images have similar characteristics to that from thermal imaging system. There is a significant component of random noise effects (from background or environments) degrading the original signal to be captured by the camera. Test conditions and processes may also induce unwanted errors or distortions into the measurement of the image quality. These include camera movement, light intensity and lens zoom. The sensitivity of some of these effects on the indexes are also presented. Some modifications of the evaluation techniques are proposed. The Modified Fidelity, MF is linearly correlated with the image quality (scattering effects) of the testing images compared to the other methods. Hence, MF is a suitable measure to quantify the scattering effect of underwater images.

Most algorithms of image processing treat an image homogeneously, while the human visual system processes a retinal image differently in the foveal and peripheral visual fields. It has been known that contrast sensitivity, spatial resolution and color discrimination of the visual system decrease from the fovea to the periphery. Moreover, recent psychophysical results showed that the spatial interactions between neighboring parts of a visual image are fundamentally different in the fovea and periphery. The perception of a visual stimulus can be suppressed or enhanced by the presence of other stimuli in its surround region. The spatial suppression in the periphery was much stronger than that in the fovea. In this report, we built an image processing model based on the neurophysiology of the human visual cortex to explore the possible impacts of the spatial interactions on image perception. We first adjusted model parameters to make the model have the same performance as human subjects had in perceiving foveal and peripheral images respectively. With those parameters, we simulated the image processing by the fovea and peripheral vision. We found that the strong spatial suppression in the periphery resulted in image boundary segmentation and salient target extraction. The response to a uniform image region was suppressed while the response to the boundary or salient regions remained. With this strategy, the visual information processed by the human cognition system is largely reduced. Based on these findings, we proposed a foveal-peripheral model for image compression and other possible applications.

The JPEG 2000 standard is a wavelet based compression methodology that achieves nearly an order of magnitude better compression performance than the existing DCT-based JPEG standard. Thus, it is desirable to convert existing JPEG images into JPEG 2000 images in order to minimize the storage and/or communications bandwidth requirements. JPEG 2000's performance enhancement can be maximized if its Region of Interest (ROI) coding option is employed. Unfortunately, this option results in a low bit rate coder, which tends to be plagued with ringing artifacts due to the abrupt truncation of high frequency wavelet coefficients at the ROI edges. In this paper, we briefly describe a novel minimum distortion transcoding technology to convert compressed JPEG images into compressed JPEG 2000 images. Experimental results indicate that the visual quality of resulting images is improved, while the time required to download them has been dramatically deceased. We expect that these new techniques can also be applied to computer vision, medical imaging, and e-commerce applications.

This paper introduces a novel adaptive cascade architecture for image compression. The idea is an extension of parallel neural network (NN) architectures which have been previously used for image compression. It is shown that the proposed technique results in higher image quality for a given compression ratio than existing NN image compression schemes. It is also shown that training of the proposed architecture is significantly faster than that of other NN-based techniques and that the number of learning parameters is small. This allows the coding process to include adaptation of the learning parameters, thus, compression does not depend on the selection of the training set as in previous single and parallel structure NN.

Bits are allocated to various subbands to minimize a particular cost function to achieve compression in subband coding. The most common cost function is the L2 norm based on mean squared error (MSE). However, the MSE often fails to correspond to the perceptual quality of the image, especially at low bit rate. In this paper, we allocate bits into various subbands by minimizing the Minkowsky metric -- a commonly used perceptual distortion measure. We then design the quantizer for each subband independent of each other based on the allocated bits. Experimental results indicate improved perceptual quality for the compressed images using Minkowsky metric compared to that of using the MSE metric.

A great deal of digital video quality measurement research has been performed to quantify human visual perception with compressed video clips. Since transmitted video quality is heavily dependent on the bit-rate, error rate and dropped packet rate, a new measurement paradigm is required to analyze the corrupted video. Fast eigen-based video quality metric (VQM) and visualization techniques have been developed to measure and analyze the corrupted video qualities objectively. 3-D SPIHT and MPEG-2 with forward error correction (FEC) had been tested over a Video CODEC RF Test-bed. The experimental results indicate that the proposed scheme is useful for a low complexity VQM.

This paper presents an error-resilient wavelet-based multiple description video coding scheme for the transmission of video over wireless channels. The proposed video coding scheme has been implemented and successfully tested over the wireless Iridium satellite communication network. As a test bed for the develope dcodec, we also present an inverse multiplexing unit that simultaneously combines several Iridium channels to form an effective higher-rate channel, where the total bandwidth is directly proportional to the number of channels combined. The developed unit can be integrated into a variety of systems such as ISR sensors, aircraft, vehicles, ships, and end user terminals (EUTs), or can operate as a standalone device. The recombination of the multi-channel unit with our proposed multi-channel video codec facilitates global and on-the-move video communications without reliance on any terrestrial or airborne infrastructure whatsoever.

In this paper, a genetic algorithm-based image compression technique using pattern classification is introduced. From one hand, the block pattern coding technique classifies image blocks into low-detailed and high-detailed blocks and codes the individual blocks according to their types. On the other hand, a genetic algorithm technique explores a given search space in parallel by means of iterative modification of a population of potenial solutions. The GA operation described here, searches for the optimal threshold(s) for the bi-level or multi level quantization of high detailed image blocks. Comparison of the results of the proposed method with the coding algorithms based on the two level minimum mean square error quantizer reveal that the former method can almost achieve optimal quantization with much less computation than required in the latter case.

This paper presents results of image interpolation with an improved method for two-dimensional cubic convolution. Convolution with a piecewise cubic is one of the most popular methods for image reconstruction, but the traditional approach uses a separable two-dimensional convolution kernel that is based on a one-dimensional derivation. The traditional, separable method is sub-optimal for the usual case of non-separable images. The improved method in this paper implements the most general non-separable, two-dimensional, piecewise-cubic interpolator with constraints for symmetry, continuity, and smoothness. The improved method of two-dimensional cubic convolution has three parameters that can be tuned to yield maximal fidelity for specific scene ensembles characterized by autocorrelation or power-spectrum. This paper illustrates examples for several scene models (a circular disk of parametric size, a square pulse with parametric rotation, and a Markov random field with parametric spatial detail) and actual images -- presenting the optimal parameters and the resulting fidelity for each model. In these examples, improved two-dimensional cubic convolution is superior to several other popular small-kernel interpolation methods.

A robust TV commercial detection system is proposed in this research. Even though several methods were investigated to address the TV commercial detection problem and interesting results were obtained before, most previous work focuses on features within a short temporal window. These methods are suitable for on-line detection, but often result in higher false alarm rates as a trade-off. To reduce the false alarm rate, we explore audiovisual features in a larger temporal window. Specifically, we group shots into scenes using audio data processing, and then obtain features that are related to commercial characteristics from scenes. Experimental results are given to demonstrate the effectiveness of the proposed system.

Our world is dominated by visual information and a tremendous amount of such information is being added day-by-day. It would be impossible to cope with this explosion of visual data, unless they are organized such that we can retrieve them efficiently and effectively. At the core of content-based image retrieval (CBIR) is the requirement that database elements must be indexed to facilitate retrieval in an efficient manner. Most existing image retrieval systems are text-based, but images frequently have little or no accompanying textual information. Problems with text-based access to images have prompted increasing interest in the development of image-based solutions. On the other hand, CBIR relies on the characterization of primitive features such as color, shape, and texture that can be automatically extracted from images themselves. Hence, the field of CBIR focuses on intuitive and efficient methods for retrieving images from a database based solely on the content contained in the images. This paper introduces a novel clustering methodology based on the gradient of images coupled with information theory (entropy) derived from statistical mechanics of "spin-up" and "spin-down" states to improve the speed of retrieval and improve the accuracy of retrieval in comparison to the traditional color histogram L1-norm retrieval methodology. By expanding the interpretation of color in images to include a gradient-based description in conjunction with information theory, a new indexing method for content-based retrieval of images from an image database is developed for the reduction of false positives in the retrieval process.

Modern imaging systems are highly sensitive in a large dynamic range, which can give images consisting of many signal levels. Presentation of such images with varying dynamic ranges on a display with a fixed number of greylevels is difficult without losing important visual information. This paper presents several algorithms for automatic dynamic range adaptation for images. All of the algorithms are suitable for and also implemented in realtime applications. They fall mainly into two categories: histogram modification techniques and frequency based techniques. Some of the algorithms are evaluated in a perception experiment, where the goal is to get the visually most attractive images, and the experiment shows that the frequency based techniques are superior to the histogram modification techniques. Some of the proposed algorithms have proven to give visually attractive images, where none or almost none of the important information is lost, for a large selection of images with varying dynamic ranges.

Image segmentation plays a crucial role in detecting cancerous lesions in breast images. Typically, the images obtained are large in dimension and it will take considerable time to run traditional image segmentation algorithms to detect and localize lesions. To increase the efficiency of the detection process, this paper develops an efficient image segmentation algorithm which limits its attention to regions where there is the possibility of lesions to exist. The image segmentation algorithm is then applied to these regions to find a threshold value. There are three primary objectives of this paper. First, to design and implement a region of interset algorithm known as the Ranking algorithm. Secondly, to identify whether the regions detected are linked using the Linkage algorithm. Thirdly, to apply the image segmentation algorithm (Otsu algorithm) to these regions to obtain a threshold value. This threshold value is then used for global image segmentation.

An Enhanced Vision System (EVS) utilizing multi-sensor image fusion is currently under development at the NASA Langley Research Center. The EVS will provide enhanced images of the flight environment to assist pilots in poor visibility conditions. Multi-spectral images obtained from a short wave infrared (SWIR), a long wave infrared (LWIR), and a color visible band CCD camera, are enhanced and fused using the Retinex algorithm. The images from the different sensors do not have a uniform data structure: the three sensors not only operate at different wavelengths, but they also have different spatial resolutions, optical fields of view (FOV), and bore-sighting inaccuracies. Thus, in order to perform image fusion, the images must first be co-registered. Image registration is the task of aligning images taken at different times, from different sensors, or from different viewpoints, so that all corresponding points in the images match. In this paper, we present two methods for registering multiple multi-spectral images. The first method performs registration using sensor specifications to match the FOVs and resolutions directly through image resampling. In the second method, registration is obtained through geometric correction based on a spatial transformation defined by user selected control points and regression analysis.

Texture analysis and synthesis methods are reviewed in this paper. Emphasis is given to recent developments in this field. As anaysis and synthesis are important in current texture based image processign and comptuer vision techniques, methods in both areas are categorized. Apart from the known categories of statistical, stuctural, model based methods, new categories are formed that include cellular automata methods, multiresolution/multiscale methods, evaolutionary methods and neural network methods. As texture is a perceptual visual cue used by the biological visual system, the biolgical basis of each method is presented. The advantages and disadvantages of each method for performgin robust texture classification are highlighted. Also, the main area of applicability of the methods and their feasibility to real-time processing are discussed. The paper gives a concise state of the art summary and analysis of texture methods used today in image processing and one can assess how close one is to the robustness and versatility of the human visual system.

The vision-based traffic monitoring system provides an attractive solution in extracting various traffic parameters such as the count, speed, flow and concentration from the processing of video data captured by a camera system. The detection accuracy is however affected by various environment factors such as shadow, occlusion, and lighting. Among these, the occurrence of occlusion is one of the major problems. In this work, a new scheme is proposed to detect the occlusion and determine the exact location of each vehicle. The proposed algorithm is based on the matching of images from multiple cameras. In the proposed scheme, we do not need edge detection, region segmentation, and camera calibration operations, which often suffer from the variation of environmental conditions. Experimental results are given to verify that the proposed technique is effective for vision-based highway surveillance systems.

The performance of error resilient MPEG video coding and transmission over wireless channels, using an XML representation called MPML, is evaluated in this research. We take into account various factors that affect the streaming performance over wireless channels. A compression algorithm is developed to reduce the MPML overhead. The error corruption model is analyzed for bitstreams with MPML protection. Error correction coding is employed to enhance the eliability of MPML header transmission. Our experimental results demonstrate the superior PSNR performance of the proposed MPML assisted video streaming system with and without the Data Partitioning (DP) mode.

Media server scheduling in video-on-demand systems includes video content allocation and request migration among servers. In this paper, we present a greedy algorithm to allocate video copies to media servers. It uses a graph model and minimizes the average shortest distance among media servers at each step. In order to study the request migration process, we introduce a state matrix representation that stores the service load information of each media server and plays an important role in the determination of migration paths. Based on this representation, we develop a state transition method to simulate the request migration process and calculate the performance metrics such as failure rates and service delay. The derived results match very well with numerical experiments. It is further demonstrated that the random early migration (REM) algorithm proposed in our previous work outperforms the normal migration scheme with lower failure rates and shorter service delay.

It is shown that the image chain has important effects upon the quality of feature extraction. Exact analytic ROC results are given for the case where arbitrary multivariate normal imagery is passed to a Bayesian feature detector designed for multivariate normal imagery with a diagonal covariance matrix. Plots are provided to allow direct visual inspection of many of the more readly apparent effects. Also shown is an analytic tradeoff that says doubling background contrast is equal to halving sensor to scene distance or sensor noise. It is also shown that the results provide a lower bound to the ROC of a Bayesian feature detector designed for arbitrary multivariate normal distributions.

The thematic mapper (TM) sensor aboard Landsats 4, 5 and enhanced TM plus (ETM+) on Landsat 7 collect imagery at 30-m sample distance in six spectral bands. New with ETM+ is a 15-m panchromatic (P) band. With image sharpening techniques, this higher resolution P data, or as an alternative, the 10-m (or 5-m) P data of the SPOT satellite, can increase the spatial resolution of the multispectral (MS) data. Sharpening requires that the lower resolution MS image be coregistered and resampled to the P data before high spatial frequency information is transferred to the MS data. For visual interpretation and machine classification tasks, it is important that the sharpened data preserve the spectral characteristics of the original low resolution data. A technique was developed for sharpening (in this case, 3:1 spatial resolution enhancement) visible spectral band data, based on a model of the sensor system point spread function (PSF) in order to maintain spectral fidelity. It combines high-pass (HP) filter sharpening methods with iterative image restoration to reduce degradations caused by sensor-system-induced blurring and resembling. Also there is a spectral fidelity requirement: sharpened MS when filtered by the modeled degradations should reproduce the low resolution source MS. Quantitative evaluation of sharpening performance was made by using simulated low resolution data generated from digital color-IR aerial photography. In comparison to the HP-filter-based sharpening method, results for the technique in this paper with simulated data show improved spectral fidelity. Preliminary results with TM 30-m visible band data sharpened with simulated 10-m panchromatic data are promising but require further study.

In practice, SAR images are very often disturbed by a correlated speckle noise. Correlation influence on wavelet coefficients is a difficult problem degrading detection and filtering performances dramatically. We establish a simple analytic relation between high order wavelet coefficient moments/cumulants and speckle second order statistics. The proposed solution is shown to improve detection and filtering algorithm performances in the presence of non white speckle noise.

In this paper, an effective method of stereo image reconstruction through the regularized adaptive disparity estimation is proposed. Although the conventional adaptive disparity estimation method can sharply improve the PSNR of a reconstructed stereo image, but some problems of overlapping between the matching windows and disallocation of the matching windows can be occurred, because the matching window size changes adaptively in accordance with the magnitude of feature values. Accordingly, in this paper, a new regularized adaptive disparity estimation technique is proposed That is, by regularizing the estimated disparity vector with the neighboring disparity vectors, problems of the conventional adaptive disparity estimation scheme might be solved, and also the predicted stereo image can be more effectively reconstructed. From some experiments using the CCETT'S stereo image pairs of 'Man' and 'Claude', it is analyzed that the proposed disparity estimation scheme can improve PSNRs of the reconstructed images to 10.89dB, 6.13dB for 'Man' and 1.41dB, 0.81dB for 'Claude' by comparing with those of the conventional pixel-based and adaptive estimation method, respectively.

This paper introduces an approach for synthesizing natural textures. Textures are modeled using a block-transition probabilistic model. In the training phase, the original textured image is split into equal size blocks, and clustered using the k-means clustering algorithm. Then, the transition probabilities between block-clusters are calculated. In the synthesis phase, the algorithm generates a sequence of indices, each representing a block-cluster, based on the transition probabilities. One advantage of this method over previous block sampling techniques is its stability. More specifically, the texture is synthesized block-by-block in a raster order. The block at a specific location is selected from one of the original image blocks. Thus, synthesis does not lead to artifacts. Additionally, the algorithm uses pre- and post- filtering. The image is filtered by a predictive filter, and the residual image is modeled using the probabilistic approach. The final synthesized image is the result of filtering the residual image by the inverse filter. Using pre- and post- processing eliminates the blockage effect. Moreover, the algorithm is computationally inexpensive, and the synthesis phase is particularly fast since it only requires generation of a sequence of cluster indices. Results show that the proposed method is successful in synthesizing realistic natural textures for a large variety of textures.

The purpose of this paper is to study the relationship between a bit map digital image and a given object, called the search object. In particular, to signal that it is likely, or not likely, that the search object appears, at least partially, in the image. Edge detection capability is assumed. Edges in the search object and in the digital image are represented as objects, in the object oriented programming sense, as Bezier cubic parameterized curves. The conversion from sequence of pixels to a Bezier polynomial representation is accomplished using least squares approximation techniques. In the event that two edges in the search object are matched with edges in the image, their relative orientation is checked using elementary vector analysis. The functioning of the algorithm is not dependent on scaling, rotation, translation, or shading of the image.

Three-dimensional (3-D) objects recognition and localization is of major importance in a wide range of applications. A number of implementations concerning the complex 3D object recognition and localization have been achieved using some heuristic approaches. However, theoretical considerations concerning the construction of invariant (or quasi invariant) relations between the types of ojbect (identifying those objects) and the position in 3D space of those objects is still a problem. That is why, a marker based method for recognition and localization of 3D objects from their 2D image is suggested in this paper. Theorems relative to the features and conditions of such markers are proposed and demonstrated. Examples are given and discussed.

In this paper, we present a new method to compress the information in an image, called MeshEZW. The proposed approach is based on the finite elements method, a mesh construction and a zerotree method. The zerotree method is an adaptive of the EZW algorithm with two new symbols for increasing the performance. These steps allow a progressive representation of the image by the automatic construction of a bitstream. The mesh structure is adapted to the image compression domain and is defined to allow video comrpession. The coder is described and some preliminary results are discussed.

Over the past few years, both large multinationals and governments have begun to contribute to even larger projects on biometric devices. Terrorist attacks in America and in other countries have highlighted the need for better identification systems for people as well as improved systems for controlling access to buildings. Another reason for investment in Research and Development in Biometric Devices, is the massive growth in internet-based systems -- whether for e-commerce, e-government or internal processes within organizations. The interface between the system and the user is routinely abused, as people have to remember many complex passwords and handle tokens of various types. In this paper an overview is given of the information that is important to know before an examination of such is systems can be done in a forensic proper way. In forensic evidence with biometric devices the forensic examiner should consider the possibilities of tampering with the biometric systems or the possibilities of unauthorized access before drawing conclusions.

This paper investigates the use of computer vision techniques to aid in the semi-automatic reconstruction of torn or ripped-up documents. First, we discuss a procedure for obtaining a digital database of a given set of paper fragments using a flatbed image scanner, a brightly coloured scanner background, and a region growing algorithm. The contour of each segmented piece of paper is then traced around using a chain code algorithm and the contours are annotated by calculating a set of feature vectors. Next, the contours of the fragments are matched against each other using the annotated feature information and a string matching algorithm. Finally, the matching results are used to reposition the paper fragments so that a jigsaw puzzle reconstruction of the document can be obtained. For each of the three major components, i.e., segmentation, matching, and global document reconstruction, we briefly discuss a set of prototype GUI tools for guiding and presenting the obtained results. We discuss the performance and the reconstruction results that can be obtained, and show that the proposed framework can offer an interesting set of tools to forensic investigators.

To reliably perform comparisons of facial images, it is important to position the head corresponding to the facial images available. Techniques using three or more landmark points on the face have been proposed for matching the face and camera positions to the available photographs. However, these methods can be cumbersome, and require the cooperation of the subject. 3D photographs, together with 3D modeling software, offer the possibility of flexible and reproducable positioning of the head of a person corresponding to the face and camera position of the facial images. We will present our experiences with a non-contact 3D laser-scanning system (Minolta VI-900), especially with respect to ease-of-use, reproducabilty, and performance for facial comparison applications.

In this study, we discussed individual camera identification of CMOS cameras, because CMOS (complementary-metal-oxide-semiconductor) imaging detectors have begun to make their move into the CCD (charge-coupled-device) fields for recent years. It can be identified whether or not the given images have been taken with the given CMOS camera by detecting the imager's intrinsic unique fixed pattern noise (FPN) just like the individual CCD camera identification method proposed by the authors. Both dark and bright pictures taken with the CMOS cameras can be identified by the method, because not only dark current in the photo detectors but also MOS-FET amplifiers incorporated in each pixel may produce pixel-to-pixel nonuniformity in sensitivity. Each pixel in CMOS detectors has the amplifier, which degrades image quality of bright images due to the nonuniformity of the amplifier gain. Two CMOS cameras were evaluated in our experiments. They were WebCamGoPlus (Creative), and EOS D30 (Canon). WebCamGoPlus is a low-priced web camera, whereas EOS D30 is for professional use. Image of a white plate were recorded with the cameras under the plate's luminance condition of 0cd/m² and 150cd/m². The recorded images were multiply integrated to reduce the random noise component. From the images of both cameras, characteristic dots patterns were observed. Some bright dots were observed in the dark images, whereas some dark dots were in the bright images. The results show that the camera identification method is also effective for CMOS cameras.

In this paper, a new watermark scheme for copyright protection of the stereo vision system using adaptive disparity matching algorithm is proposed. That is, the right image of a stereo image is formulated to 8×8 blocks, in which each block is composed of 64×64 and the DCT coefficients for each of these blocks are computed and then, a specific watermark image having 64×64 pixels is embedded into these DCT coefficients. After disparity information is extracted from both of the watermarked right image and left image using a matching algorithm, the left image and disparity information is transmitted to the recipient through the communication channel. Then, the watermark is extracted from the reconstructed stereo image using the same seed value used in the embedding process. The watermark is extracted from the reconstructed right image as a result extraction performance of a watermark tightly depends on the employed disparity matching algorithms. From some experiments using a sequential stereo image of 'Fichier' and the watermark image of English alphabet, it is shown that PSNR of the extracted watermark image from the reconstructed image by using an adaptive disparity matching algorithm improves to 3.68 dB on average by comparing with those of the pixel and block-matching algorithms.