Image sensors for digital cameras are built with ever decreasing pixel sizes. The size of the pixels seems to be limited by technology only. However, there is also a hard theoretical limit for classical video camera systems: During a certain exposure time only a certain number of photons will reach the sensor. The resulting shot noise thus limits the signal-to-noise ratio. In this letter we show that current sensors are already surprisingly close to this limit.

Abstract unavailable.

In the case of textured images and more particularly of directional textures, a new parametric technique is proposed to estimate the orientation field of textures. It consists of segmenting the image into regions with homogeneous orientations, and estimating the orientation inside each of these regions. This allows us to maximize the size of the samples used to estimate the orientation without being corrupted by the presence of boundaries between regions. For that purpose, the local-hence noisy-orientations of the texture are first estimated using small filters (3×3  pixels). The segmentation of the obtained orientation field image then relies on a generalization of a minimum description length based segmentation technique, to the case of π-periodic circular data modeled with Von Mises probability density functions. This leads to a fast segmentation algorithm without tuning parameters in the optimized criterion. The accuracy of the orientations estimated with the proposed method is then compared with other approaches on synthetic images and an application to the processing of real images is finally addressed.

Fabric's smoothness is a key factor in determining the quality of finished textile products and has great influence on the functionality of industrial textiles and high-end textile products. With popularization of the zero defect industrial concept, identifying and measuring defective material in the early stage of production is of great interest to the industry. In the current market, many systems are able to achieve automatic monitoring and control of fabric, paper, and nonwoven material during the entire production process, however online measurement of hairiness is still an open topic and highly desirable for industrial applications. We propose a computer vision approach to compute epipole by using variable homography, which can be used to measure emergent fiber length on textile fabrics. The main challenges addressed in this paper are the application of variable homography on textile monitoring and measurement, as well as the accuracy of the estimated calculation. We propose that a fibrous structure can be considered as a two-layer structure, and then we show how variable homography combined with epipolar geometry can estimate the length of the fiber defects. Simulations are carried out to show the effectiveness of this method. The true length of selected fibers is measured precisely using a digital optical microscope, and then the same fibers are tested by our method. Our experimental results suggest that smoothness monitored by variable homography is an accurate and robust method of quality control for important industrial fabrics.

The ability to determine the size and shape of a room or other interior space prior to the entry of personnel can identify risks, aid in planning, and promote efficient allocation of resources. The performance of infrared sensors, acoustic sensors, frequency modulated continuous wave radar, and stereo cameras for mapping an enclosed space is evaluated. A multisensor system and a communication network are demonstrated to map accurately the interior of a building and generate a consistent solution and robust results. The average error of the system is 10 to 30 cm. Processing time for the data fusion algorithm is typically 5 to 20 sec/room to generate the floor plan for the building. Two practical examples are presented. Current limitations of the technology are discussed, and approaches to abate them are proposed.

Fashion and design greatly influence the conception of manufactured products which now exhibit complex forms and shapes. Two-dimensional quality control procedures (e.g., shape, textures, colors, and 2D geometry) are progressively being replaced by 3D inspection methods (e.g., 3D geometry, colors, and texture on the 3D shape) therefore requiring a digitization of the object surface. Three dimensional surface acquisition is a topic which has been studied to a large extent, and a significant number of techniques for acquiring 3D shapes has been proposed, leading to a wide range of commercial solutions available on the market. These systems cover a wide range from micro-scale objects such as shape from focus and shape from defocus techniques, to several meter sized objects (time of flight technique). Nevertheless, the use of such systems still encounters difficulties when dealing with non-diffuse (non Lambertian) surfaces as is the case for transparent, semi-transparent, or highly reflective materials (e.g., glass, crystals, plastics, and shiny metals). We review and compare various systems and approaches which were recently developed for 3D digitization of transparent objects.

Textiles are mainly used for decoration and protection. In both cases, their original appearance and its retention are important factors for customers. Therefore, evaluation of appearance parameters are critical for quality assurance purposes, during and after manufacturing, to determine the lifetime and/or beauty of textile products. In particular, appearance retention of textile products is commonly certified with grades, which are currently assigned by human experts. However, manufacturers would prefer a more objective system. We present an objective system for grading appearance retention, particularly, for textile floor coverings. Changes in appearance are quantified by using linear regression models on texture features extracted from intensity and range images. Range images are obtained by our own laser scanner, reconstructing the carpet surface using two methods that have been previously presented. We extract texture features using a variant of the local binary pattern technique based on detecting those patterns whose frequencies are related to the appearance retention grades. We test models for eight types of carpets. Results show that the proposed approach describes the degree of wear with a precision within the range allowed to human inspectors by international standards. The methodology followed in this experiment has been designed to be general for evaluating global deviation of texture in other types of textiles, as well as other surface

As a single-shot interferometric technique, spatial carrier interferometry has been thoroughly investigated, and it has been shown to have some problems, such as low spatial resolution. To overcome the problems, we propose a novel single-shot surface profiling technique that does not require carrier introduction. It is based on a model-fitting algorithm and estimates the model parameters and the heights of plural points simultaneously based on their multiwavelength intensity data. The validity of the proposed method is demonstrated by computer simulations and actual experiments.

Many workpieces produced in large numbers with a large variety of sizes and geometries, e.g. castings and forgings, have to be 100% inspected. In addition to geometric tolerances, material defects, e.g. surface cracks, also have to be detected. We present a fully automated nondestructive testing technique for both types of defects. The workpiece is subject to continuous motion, and during this motion two measurements are performed. In the first step, after applying a short inductive heating, a thermographic measurement is carried out. An infrared camera records the surface temperature of the workpiece enabling the localization of material defects and surface cracks. In the second step, a light sectioning measurement is performed to measure the three-dimensional geometry of the piece. With the help of feature-based registration the data from the two different sources are fused and evaluated together. The advantage of this technique is that a more reliable decision can be made about the nature of the failures and their possible causes. The same registration technique also can be used for the comparison of different pieces and therefore to localize different failure types, via comparison with a "golden," defect-free piece. The registration technique can be applied to any part that has unique geometric features, around which moments can be computed. Consequently, the inspection technique can be applied to many different parts. The efficacy of the method is demonstrated with measurements on three parts having different geometries.

Current algorithms for automated processing of Vickers hardness testing images are unsuitable for a broad range of images that are taken in industrial environments because such images show great variations in the Vickers indentation as well as in the specimen surface. The authors present a three-stage multiresolution template matching algorithm that shows excellent results, even for such challenging images. The capabilities of this algorithm are compared to known algorithms from the literature and results are presented. The comparison is conducted on two significant indentation image databases with 150 and 216 highly varying images. The applicability of the proposed algorithm is further illustrated by its competitive runtime performance.

The progress in image acquisition techniques provides life sciences with an abundance of data. Image analysis facilitates the assessment. The actin cytoskeleton plays a crucial role in understanding the behavior of osteoblastic cells on biomaterials. In the flat basal part of the cells, it can be visualized by confocal laser scanning microscopy. In the microscopic images, the stained cytoskeleton appears as a dense network of bright ridges which is so far only qualitatively assessed. For its quantification, there is a need for ridge detection techniques that provide a geometrical description of this graph feature. The state of the art methods do not cope with the systematical degradation by noise, unspecific luminance, and uneven dye uptake. This work presents the key part of a ridge-tracking technique, which makes more efficient use of context information, and evaluate it by its length measurement accuracy. Two random models illustrate the performance against ground truth. Representative microscopic images confirm the applicability.

The three dimensional (3D) X-ray computed tomography (3D-CT) has proven its successful application as an inspection method in nondestructive testing. The generated 3D volume uses high efficiency reconstruction algorithms containing all required information on the inner structures of the inspected part. Segmentation of this volume reveals suspicious regions that need to be classified as defective or false alarms. This paper deals with the classification step using data fusion theory, which was successfully applied on 2D X-ray data in previous work along with a support vector machine (SVM). For this study we chose a 3D-CT dataset of aluminium castings that needs to be fully inspected via X-ray CT to ensure their quality. We achieved a true classification rate of 97% on a validation dataset, which proves the effectiveness of the data fusion theory as a method to build a better classifier. Comparison with SVMs shows the importance of selecting the most pertinent features to improve the classifier performance and attaining 98% of true classification rate.

The goal of image fusion is to combine multiple source images obtained using different capture techniques into a single image to provide an effective contextual enhancement of a scene for human or machine perception. In practice, considerable value can be gained in the fusion of images that are dissimilar or complementary in nature. However, in such cases, global weighting schemes may not sufficiently weigh the contribution of the pertinent information of the source images, while existing adaptive schemes calculate weights based on the relative amounts of salient features, which can cause severe artifacting or inadequate local luminance in the fusion result. Accordingly, a new multiscale image fusion algorithm is proposed. The approximation coefficient fusion rule of the algorithm is based on a novel similarity based weighting scheme capable of providing improved fusion results when the input source images are either similar or dissimilar to each other. Moreover, the algorithm employs a new detail coefficient fusion rule integrating a parametric multiscale contrast measure. The parametric nature of the contrast measure allows the degree to which psychophysical laws of human vision hold to be tuned based on image-dependent characteristics. Experimental results illustrate the superior performance of the proposed algorithm qualitatively and quantitatively.

Accurate quantitative reconstruction in kV cone-beam computed tomography (CBCT) is challenged by the presence of secondary radiations (scattering, fluorescence, and bremsstrahlung photons) coming from the object and from the detector itself. The authors present a simulation study of the CBCT imaging chain and its integration into a comprehensive correction algorithm. A layer model of the flat-panel detector is built in a Monte Carlo environment in order to help in localizing and analyzing the secondary radiations. The contribution of these events to the final image is estimated with a convolution model to account for detector secondary radiations combined with a forced-detection scheme to speed-up the Monte Carlo simulation without loss of accuracy. We more specifically assess to what extent a 2D description of the flat-panel detector would be sufficient for the forward model (i.e., the image formation process) of an iterative correction algorithm, both in terms of energy and incidence angle of incoming photons. Results show that both object and detector secondary radiations have to be considered in CBCT. The correction algorithm iteratively compensates for the secondary radiations and the beam hardening in object space. Preliminary results on tomographic acquisitions demonstrate a quantitative improvement on the first iteration.

A large variety of computationally lightweight functions used for assessing image sharpness in the spatial domain is evaluated for application in a passive autofocus system in the context of microindentation-based Vickers hardness testing. Alternatively, the file size of compressed JPEG images is proposed to determine image sharpness. The functions are evaluated on a significant dataset of microindentation images with respect to their properties required for focus search and sharpness assessment, their robustness to downsampling the image data, and their computational demand. Experiments suggest that of all spatial domain techniques considered, the simple Brenner autofocus function is the best compromise between accuracy and computational effort, while JPEG file size is a versatile solution when the application context allows.

Quantification of the overlapping particles in crystallization processes is very important for the quality control of chemical products or drugs. We present a method of segmentation of polygonal-shaped (i.e., rectangles, regular/irregular prisms) and overlapping particles from in situ images during a crystallization process for measuring their size distributions. The method is first based on detecting the geometric features of the particles identified by their salient corners. A clustering technique is then applied by grouping three correspondent salient corners belonging to the same particle. The proposed method is applied on particles of ammonium oxalate during batch crystallization in pure water. The particle size distributions are calculated, and a quantitative comparison between the proposed method and a manual sizing is performed. The method showed that it is valid for analyzing the crystal growth, and the results are promising for monitoring the particle size distribution.

We address the challenge of parallelization development of industrial high-performance inspection systems comparing a conventional parallelization approach versus an auto-parallelized technique. Therefore, we introduce the functional array processing language Single Assignment C (SAC), which relies on a hardware virtualization concept for automated, parallel machine code generation for multi-core CPUs and GPUs. Additional software engineering aspects like programmability, productivity, understandability, maintainability, and resulting achieved gain in performance are discussed from the point of view of a developer. With several illustrative benchmarking examples from the field of image processing and machine learning, the relationship between runtime performance and efficiency of development is analyzed.

This paper presents a new technique for contrast enhancement for color images called histogram shifting with alpha rooting. The novelty in the presented method consists in adapting spatial domain techniques into the transform domain. The benefits of operating in the transform domain include low complexity of computations, ease of viewing, and manipulation of the frequency composition of the image and preservation of the phase information. The combination of the alpha-rooting algorithm, coupled with histogram shifting shows the method's effectiveness for enhancing overexposed images. The contrast enhancement parameter of the algorithm is established automatically based on the entropy of the images. A comprehensive comparative study on image-enhancement algorithms based on discrete cosine transform coefficients is provided. Computer simulations and analysis are provided to compare the enhancement performance of the proposed technique to state of the art approaches. We perform a statistical analysis on the results and quantitatively show that the proposed approach performs well for color image enhancement, which is also validated by ratings from human observers.

In human retina observation (with non mydriatic optical microscopes), an image registration process is often employed to enlarge the field of view. Analyzing all the images takes a lot of time. Numerous techniques have been proposed to perform the registration process. Its good evaluation is a difficult question that is then raising. This article presents the use of two quantitative criterions to evaluate and compare some classical feature-based image registration techniques. The images are first segmented and the resulting binary images are then registered. The good quality of the registration process is evaluated with a normalized criterion based on the ε dissimilarity criterion, and the figure of merit criterion (fom), for 25 pairs of images with a manual selection of control points. These criterions are normalized by the results of the affine method (considered as the most simple method). Then, for each pair, the influence of the number of points used to perform the registration is evaluated.

Color correction is important in vision applications. We aim to improve conventional transformation with a linear hypothesis of color correction (i.e., von Kries method) by innovatively using a hue-determined matrix. A new model, named the hue-compensated diagonal model (HCDM), is proposed to benefit correcting colors with chromatic-saturated illuminations. Inspired by the chromatic adaptation occurring in the human visual system (HVS), the HCDM reinforces the illuminant complementary color by weighting the von Kries coefficients. These HCDM weighting terms, being illuminant-hue specific based on the relative absorption model (RAM), consist of the proposed hue-chromatic characteristic function with learning parameters. As a result, the HCDM out-performed both the principle component analysis (PCA) approach and the von Kries method in most synthetic experiments, and achieved better results than the von Kries method in real image experiments. Thus, the proposed HCDM, with global-processing capability and real-image applicability, can be used as an effective model in color reproductions and image applications.

A simple and practical technique is presented for creating fine three-dimensional (3D) images with polygon-based computer-generated holograms. The polygon-based method is a technique for computing the optical wave-field of virtual 3D scenes given by a numerical model. The presented method takes less computation time than common point-source methods and produces fine spatial 3D images of deep 3D scenes that convey a strong sensation of depth, unlike conventional 3D systems providing only binocular disparity. However, smooth surfaces cannot be reconstructed using the presented method because the surfaces are approximated by planar polygons. This problem is resolved by introducing a simple rendering technique that is almost the same as that in common computer graphics, since the polygon-based method has similarity to rendering techniques in computer graphics. Two actual computer holograms are presented to verify and demonstrate the proposed technique. One is a hologram of a live face whose shape is measured using a 3D laser scanner that outputs polygon-mesh data. The other is for a scene including the moon. Both are created employing the proposed rendering techniques of the texture mapping of real photographs and smooth shading.

We propose a geometrical approach for virtual camera rectification on uncalibrated single-lens stereovision using a biprism. This system is also called a virtual stereovision system, as the image captured can be divided into two which are equivalent to two images captured using two cameras with different perspectives. The proposed method is divided into two parts. The first part is to compute the projection transformation matrix of two virtual cameras based on a unique geometrical ray sketching, which can accurately recover the extrinsic parameters, and the second part is to compute the rectification transformation matrix, which is applied on the images captured using the system. As the geometrical analysis eliminates the complex calibration process and rectification reduces the correspondence searching to one-dimensional, this method provides a simple stereo matching technique for this system. Experimental results are presented to show the effectiveness of the approach.

We present spatiotemporal denoising based on overlapped motion compensation and advanced collaborative filtering. First, noise-robust overlapped motion compensation is performed on a block basis for temporal grouping. Next, the K-nearest neighbors of each block are grouped in a 3D array, and the 3D array is transformed. Then, adaptive soft thresholding is performed in the 3D transform domain. In addition, a modified weighting strategy for aggregation is applied for better visual quality. Simulation results show that the proposed algorithm improves the peak signal-to-noise ratio performance by about 2 dB in comparison with the state-of-the-art technique while providing much better subjective visual quality.

A novel adaptive high-performance and nonlinear filtering algorithm (AHPNFA) is proposed for the removal of mixed noise from corrupted images. This algorithm contains two stages: noise determining and noise removing. In the first stage, for each filtering window of the processed image, the center pixel and its nearest neighbors are modeled as a statistical variable. By exploiting Chebyshev's theorem, the fuzzy mean process is used to estimate adaptively the detection parameters that are required in determining whether the current pixel is corrupted or not. In the second stage, the Radon transform is performed to determine the texture direction probability density distributions of local areas of images, then using the texture direction probability density distributions and the local features of the image remove the noisy pixels. To demonstrate the advantages, the proposed AHPNFA is compared with the other test filters both visually and quantitatively. Simulation results show that the proposed AHPNFA has better values in important evaluation metrics; in particular, the computational complexity of the AHPNFA is about five to 11 times lower than the other test filters used in the comparison, therefore, the proposed AHPNFA has better performances.

Following the recently proposed total variation (TV) up-sampling method, we present a novel image up-sampling algorithm based on fractional-order bidirectional diffusion. For the bidirectional diffusion, the forward diffusion occurs on the light side of edge, while the backward diffusion proceeds on the dark side. This bidirectional diffusion can reduce the edge width and avoid the appearances of false edge or texture and block effects. Moreover, the fractional-order derivative is used to avoid strong contrast near the edge in the interpolated images. The experiments show that, unlike the TV up-sampling (TVUP) method, the proposed algorithm does not suffer from the drift of edges, block effect in the smooth regions, false edges, and false texture.

A highly efficient postprocessing technique which enables the result of edge detection to be used for image segmentation is proposed. The method starts from an edge map obtained by a standard edge detection tool, e.g., Canny edge detector, and corrects it to obtain an edge map in which every edge point belongs to a closed boundary of an image region. The correction of the original edge map assumes removing some of the existing edge points as well as inserting virtual edge points. The proposed edge map correction procedure consists of two stages: (1) edge linking, which closes the gaps in edge contours by inserting virtual edge elements, and (2) edge pruning, which rejects spurious contours thereby avoiding over-segmentation. The edge pruning procedure performs an iterative greedy minimization of a correction cost function, while keeping all contours of the edge map closed. The proposed approach is evaluated using a set of standard test images.

Use of multispectral magnetic resonance imaging has received a great interest for prostate cancer localization in research and clinical studies. Manual extraction of prostate tumors from multispectral magnetic resonance imaging is inefficient and subjective, while automated segmentation is objective and reproducible. For supervised, automated segmentation approaches, learning is essential to obtain the information from training dataset. However, in this procedure, all patients are assumed to have similar properties for the tumor and normal tissues, and the segmentation performance suffers since the variations across patients are ignored. To conquer this difficulty, we propose a new iterative normalization method based on relative intensity values of tumor and normal tissues to normalize multispectral magnetic resonance images and improve segmentation performance. The idea of relative intensity mimics the manual segmentation performed by human readers, who compare the contrast between regions without knowing the actual intensity values. We compare the segmentation performance of the proposed method with that of z-score normalization followed by support vector machine, local active contours, and fuzzy Markov random field. Our experimental results demonstrate that our method outperforms the three other state-of-the-art algorithms, and was found to have specificity of 0.73, sensitivity of 0.69, and accuracy of 0.79, significantly better than alternative methods.

Block sizes larger than the traditional 16×16 macroblock (MB) structure and extended macroblock (EMB) size motion estimation (ME) are introduced in joint model key technology areas software to improve the inter block coding efficiency. Similar to H.264, variable size motion estimation (VS-ME) is still adopted for each 16×16 MB including 16×16, 16×8, 8×16, 8×8, 8×4, 4×8, and 4×4. EMB size ME and VS-ME achieve the highest possible coding efficiency, but result in extremely large coding time which obstructs it from practical use. A fast mode decision algorithm for EMBs based on motion homogeneity is proposed to reduce the computational complexity of ME. The basic idea of the proposed method is to utilize the spatial property of motion field in prediction where EMB size ME and small size ME (including 8×8, 8×4, 4×8 and 4×4) are needed, and only in these regions EMB size ME and small size ME are enabled. The motion field is generated by the corresponding motion vectors in spatial window. Simulation results show that the proposed algorithm can save 55% average computational complexity, with negligible loss of coding efficiency.

We propose a new method for human action recognition based on multiple features and a hybrid generative/discriminative model. Specifically, we propose a new action representation based on computing a rich set of descriptors from Affine-SIFT key point trajectories. A new hybrid generative/discriminative approach based on support vector machine and topic model is proposed using Fisher kernel method for action recognition. Fisher score for the topic model is evaluated by the variational inference algorithm. To obtain efficient and compact representations for actions, we develop a feature fusion method to combine spatial-temporal local motion descriptors and demonstrate how this kernel framework can be used to combine different types of features and models into a single classifier. Our experiments, conducted on a number of popular datasets, show performance improvements over the corresponding generative approach and are competitive with the best results reported in the literature.

A novel method for three-dimensional (3-D) shape retrieval using bag-of-feature techniques (BoF) is proposed. This method is based on vector quantization of invariant descriptors of 3-D object patches. Firstly, it starts by selecting and then describing a set of points from the 3-D object. Such descriptors have the advantage of being invariant to different transformations that a shape can undergo. Based on vector quantization, we cluster those descriptors to form a shape vocabulary. Then, each point selected in the object is associated to a cluster (word) in that vocabulary. Finally, a weighted vector counting the occurrences of every word is computed. These results clearly demonstrate that the method is robust to nonrigid and deformable shapes, in which the class of transformations may be very wide due to the capability of such shapes to bend and assume different forms.

Time-of-flight (TOF) full-field range cameras use a correlative imaging technique to generate three-dimensional measurements of the environment. Though reliable and cheap they have the disadvantage of high measurement noise and errors that limit the practical use of these cameras in industrial applications. We show how some of these limitations can be overcome with standard image processing techniques specially adapted to TOF camera data. Additional information in the multimodal images recorded in this setting, and not available in standard image processing settings, can be used to improve reduction of measurement noise. Three extensions of standard techniques, wavelet thresholding, adaptive smoothing on a clustering based image segmentation, and an extended anisotropic diffusion filtering, make use of this information and are compared on synthetic data and on data acquired from two different off-the-shelf TOF cameras. Of these methods, the adapted anisotropic diffusion technique gives best results, and is implementable to perform in real time using current graphics processing unit (GPU) hardware. Like traditional anisotropic diffusion, it requires some parameter adaptation to the scene characteristics, but allows for low visualization delay and improved visualization of moving objects by avoiding long averaging periods when compared to traditional TOF image denoising.

A novel method of lossy compression for images of text documents is proposed. The method is based on classifying the objects, characters, and pictures that appear in the images. We used the Tanimoto distance to group the objects into different classes to create an object dictionary; then, we codified the instances of each class by means of a code of three symbols called the three orthogonal symbol chain code (3OT). We applied an entropy coder to the resulting chain, which groups the symbols of 3OT; finally, we compressed the chain obtained by using the Paq8l archiver, which is based on a context-mixing algorithm divided into a predictor and an arithmetic coder. We obtained a high performance in memory storage, with an average of 2.17 times better compression levels with respect to the international standard Joint Bi-level Image Experts Group 2 on its lossy information version.

The adaptive finite element method (AFEM) is an effective way to retain enough resolution in selected regions while improving computational efficiency for reconstruction in fluorescent molecular tomography (FMT). In addition, background optical properties acquired by diffuse optical tomography (DOT) will improve the quality of reconstructed images. In this work, a DOT-guided adaptive reconstruction method (DGARM) is proposed, in which DOT is introduced into an adaptive reconstruction framework of FMT. Besides being used as functional a priori information in the forward modeling, the reconstructed local absorption coefficients in DOT are also used to form internal structural a priori information, which is further used to identify regions of interest (ROIs) in the mesh refinement of the adaptive procedure. Because optical properties are recovered before the formulation of FMT, a linear relationship is established between the fluorescent field and boundary measurements. We implemented numerical simulations and physical experiments to evaluate the performance of the algorithm. Compared with the strategies of uniform meshing with DOT and AFEM without DOT, DGAM improves localization accuracy of the reconstruction of fluorescence inclusion.

In many tasks of machine vision applications, it is important that recorded colors remain constant, in the real world scene, even under changes of the illuminants and the cameras. Contrary to the human vision system, a machine vision system exhibits inadequate adaptability to the variation of lighting conditions. Automatic white balance control available in commercial cameras is not sufficient to provide reproducible color classification. We address this problem of color constancy on a large image database acquired with varying digital cameras and lighting conditions. A device-independent color representation may be obtained by applying a chromatic adaptation transform, from a calibrated color checker pattern included in the field of view. Instead of using the standard Macbeth color checker, we suggest selecting judicious colors to design a customized pattern from contextual information. A comparative study demonstrates that this approach ensures a stronger constancy of the colors-of-interest before vision control thus enabling a wide variety of applications.

Successful execution of tasks such as image classification, object detection and recognition, and scene classification depends on the definition of a set of features able to describe images effectively. Texture is among the features used by the human visual system. It provides information regarding spatial distribution, changes in brightness, and description regarding the structural arrangement of surfaces. However, although the visual human system is extremely accurate to recognize and describe textures, it is difficult to define a set of textural descriptors to be used in image analysis on different application domains. This work evaluates several texture descriptors and demonstrates that the combination of descriptors can improve the performance of texture classification.

We propose an audio-visual approach to video genre classification using content descriptors that exploit audio, color, temporal, and contour information. Audio information is extracted at block-level, which has the advantage of capturing local temporal information. At the temporal structure level, we consider action content in relation to human perception. Color perception is quantified using statistics of color distribution, elementary hues, color properties, and relationships between colors. Further, we compute statistics of contour geometry and relationships. The main contribution of our work lies in harnessing the descriptive power of the combination of these descriptors in genre classification. Validation was carried out on over 91 h of video footage encompassing 7 common video genres, yielding average precision and recall ratios of 87% to 100% and 77% to 100%, respectively, and an overall average correct classification of up to 97%. Also, experimental comparison as part of the MediaEval 2011 benchmarking campaign demonstrated the efficiency of the proposed audio-visual descriptors over other existing approaches. Finally, we discuss a 3-D video browsing platform that displays movies using feature-based coordinates and thus regroups them according to genre.

The contours of benign masses and malignant tumors in mammograms and other types of breast images differ substantially in their shape and complexity; the former are usually round and smooth, whereas the latter are typically spiculated and irregular. We demonstrate the usefulness of fractal analysis via a frequency domain approach applied to one-dimensional signatures of the two-dimensional contours of breast masses in mammograms. The 1/f model related to fractional Brownian motion was applied via power spectral analysis of signatures to estimate the fractal dimension. Experiments with a dataset of 111 contours, including those of 65 benign masses and 46 malignant tumors, indicated a high classification performance of 0.8962 in terms of the area under the receiver-operating characteristic curve. The method should be useful in computer-aided diagnosis of breast cancer.

We propose to use the codispersion coefficient to define a measure of similarity between images. This coefficient has been widely used in spatial statistics to quantify the association between two spatial processes, and here we explore its capabilities in an image processing context is mathematically simple to compute and possesses good statistical properties. The new measure takes into account the spatial association in a specific direction h between a degraded image and the original unmodified image. Three applications are developed to illustrate the capabilities of our proposal. The defined measure captures the spatial association produced by fitting AR-2D processes with different window sizes. It is able to distinguish the levels of similarity between two images for specific directions in two-dimensional space. Finally, it detects stochastic resonance when an image is transmitted by a nonlinear device.

We discuss improved image reconstruction and segmentation in a framework we term model-controlled flooding (MCF). This extends the watershed transform for segmentation by allowing the integration of a priori information about image objects into flooding simulation processes. Modeling the initial seeding, region growing, and stopping rules of the watershed flooding process allows users to customize the simulation with user-defined or default model functions incorporating prior information. It also extends a more general class of transforms based on connected attribute filters by allowing the modification of connected components of a grayscale image, thus providing more flexibility in image reconstruction. MCF reconstruction defines images with desirable features for further segmentation using existing methods and can lead to substantial improvements. We demonstrate the MCF framework using a size transform that extends grayscale area opening and attribute thickening/thinning, and give examples from several areas: concealed object detection, speckle counting in biological single cell studies, and analyses of benchmark microscopic image data sets. MCF achieves benchmark error rates well below those reported in the recent literature and in comparison with other algorithms, while being easily adapted to new imaging contexts.

Spatial filtering, which aims to mimic the contrast sensitivity function (CSF) of the human visual system (HVS), has previously been combined with color difference formulae for measuring color image reproduction errors. These spatial filters attenuate imperceptible information in images, unfortunately including high frequency edges, which are believed to be crucial in the process of scene analysis by the HVS. The adaptive bilateral filter represents a novel approach, which avoids the undesirable loss of edge information introduced by CSF-based filtering. The bilateral filter employs two Gaussian smoothing filters in different domains, i.e., spatial domain and intensity domain. We propose a method to decide the parameters, which are designed to be adaptive to the corresponding viewing conditions, and the quantity and homogeneity of information contained in an image. Experiments and discussions are given to support the proposal. A series of perceptual experiments were conducted to evaluate the performance of our approach. The experimental sample images were reproduced with variations in six image attributes: lightness, chroma, hue, compression, noise, and sharpness/blurriness. The Pearson's correlation values between the model-predicted image difference and the observed difference were employed to evaluate the performance, and compare it with that of spatial CIELAB and image appearance model.

Efficient utilization of bandwidth is key to providing mobile multimedia services such as wireless imaging. As such, efficient allocation of the available parity budget for error protection of the associated codestreams is essential for ensuring quality of service. Computing the optimal parity allocation to the codestream packets may not be possible in real-time due to the huge number of ways the parity can be assigned to the codestream packets. This is particularly true for systems having limited resources, such as mobile handheld devices. As a result, it is important to provide an error-protection scheme that gives good error-protection performance while imposing low computational and memory demands on the system. We therefore analyze how different parameters such as signal-to-noise ratio and source distortion-rate function affect the optimal unequal error protection (UEP) scheme for embedded codestreams. We also propose to approximate the optimal UEP using suitable mathematical models in order to reduce the complexity that otherwise would occur with computing the optimal UEP. We investigate several models for their fitness in approximating the optimal UEP using different performance metrics. The simulation results show that most of these models provide an excellent trade-off between performance and complexity.

We discuss the problem of recognizing the shape of planar objects consisting of "blobs" that can be modeled as Gaussian mixture densities. We describe an empirical comparison method, assuming a large number of independent samples are given for each distribution. Instead of comparing the Gaussian mixtures directly, we compare the underlying distribution of distances of each mixture. Since distances are invariant under rotations and translations, this provides a workaround to the problem of aligning the objects before comparing them-thus speeding the comparison process. We prove that the distribution of distances is a lossless representation of the shape of generic Gaussian mixtures. Our numerical experiments indicate that, when all the components of the Gaussian mixtures are equally weighted and have the same standard deviation matrix, the proposed method is no less accurate than methods that compare the planar mixtures directly. The extension of our method to the problem of recognizing halftone patterns is briefly discussed.

We present two semi-automatic solution methods for the three dimensional (3D) segmentation of cavity and myocardium from a 3D cardiac multislice CT (MSCT) data. The main framework of the segmentation algorithms is based on random walks, in which the novelty lies in a seeds-selection method composed of region growing technique and morphological operation to locate and identify the cavity and myocardium of the left ventricle (LV). In the first solution, a semi-automatic segmentation approach (Method_1) is suggested to extract the epi-cardium and endo-cardium boundaries of LV of the heart. This proposed solution depends on the use of the normal random walker algorithm. In the second solution, a semi-automatic segmentation approach (Method_2) based on improved random walks is proposed. Segmentation is done within the framework of toboggan algorithm in combination with a random walk based technique. The two proposed semi-automatic segmentation methods either based on the normal random walker or the improved random walker algorithms utilized six-connected lattice topology and a conjugate gradient method to promote the segmentation performance of the 3D data. The two semi-automatic solution methods were evaluated using 20 cardiac MSCT datasets. Semi-automatic generated contours were compared to expert contours. For Method_1, 83.4% of epi-cardial contours and 74.7% of endo-cardial contours had a maximum error of 5 mm along 95% of the contour arc length. For Method_2, those numbers were 94.3% (epi-cardium) and 92.3% (endo-cardium), respectively. Volume regression analysis revealed good linear correlations between manual and semiautomatic volumes, r ≥ 0.99.

We discuss methods for modeling and removal of noise in astronomical images. For its favorable properties, we exploit the undecimated wavelet representation and apply noise suppression in this domain. Usually, the noise analysis of the studied imaging system is carried out in the spatial domain. However, noise in astronomical data is non-Gaussian, and thus the noise model parameters need to be estimated directly in the wavelet domain.We derive equations for estimating the sample moments for non-Gaussian noise in the wavelet domain. We consider that the sample moments in the spatial domain are known from the noise analysis and that the model parameters are estimated by using the method of moments.

We develop a new algorithm that extends the bidimensional fast digital radon transform from Götz and Druckmüller (1996) to digitally simulate the refocusing of a 4-D lightfield into a 3-D volume of photographic planes as previously done by Ng et al. (2005) but with the minimum number of operations. This new algorithm does not require multiplications just sums and its computational complexity is O(N⁴) to achieve a volume consisting of 2N photographic planes focused at different depths from a N⁴ plenoptic image. This reduced complexity allows for the acquisition and processing of a plenoptic sequence with the purpose of estimating 3-D shape at video rate. Examples are given of implementations on GPU and CPU platforms. Finally, a modified version of the algorithm to deal with domains of sizes different than a power of two is proposed.

We propose a colorization algorithm for grayscale images and videos that demands simple user scribbles on a selected set of source pixels. The proposed algorithm colorizes non source pixels by propagating the colors of source pixels. To achieve reliable colorization results, we adopt a prioritization approach, which colorizes large smooth regions first and small detailed regions later using a Gaussian pyramid of gradient images. At each level of the pyramid, we determine the priority of each non source pixel using the luminance similarities to neighboring pixels and the geometric proximities to source pixels. Then we repeatedly colorize the pixel with the highest priority by interpolating the colors of neighboring source pixels. Also, for video colorization, the proposed algorithm transfers the colors of the first frame to subsequent frames based on optical flow estimation. To correct incorrectly transferred colors, the accuracies of transferred colors are calculated, and inaccurate colors are refined using neighboring colors. Simulation results show that the proposed algorithm provides accurate and reliable colorization results. We also demonstrate that the proposed algorithm can be used for color restoration and recolorization.

We propose a double bilateral (DoBi) filter that consists of a classical bilateral filter and a new bilateral filter for image restoration. Bilateral filtering is a simple, noniterative, and effective denoising filter that smooths images while preserving edges by means of a nonlinear combination of adjacent pixel values. A median-metric weighting function is introduced by incorporating a switching median filter into the similarity function. This median-metric component associated with a spatial function constitute the second bilateral filter, which compensates the classical bilateral filter. Moreover, a parameter automation mechanism is proposed to facilitate the restoration procedure. A wide variety of images contaminated by various degrees of Gaussian, impulse, and mixed noise were used to assess the performance of this new restoration algorithm. Experimental results indicated that the DoBi filter outperformed several state-of-the-art methods in both visual image quality and restored signal quantity.

Contemporary research in automated panorama creation utilizes camera calibration or extensive knowledge of camera locations and relations to each other to achieve successful results. Research in image registration attempts to restrict these same camera parameters or apply complex point-matching schemes to overcome the complications found in real-world scenarios. This paper presents a novel automated panorama creation algorithm by developing an affine transformation search based on maximized mutual information (MMI) for region-based registration. Standard MMI techniques have been limited to applications with airborne/satellite imagery or medical images. We show that a novel MMI algorithm can approximate an accurate registration between views of realistic scenes of varying depth distortion. The proposed algorithm has been developed using stationary, color, surveillance video data for a scenario with no a priori camera-to-camera parameters. This algorithm is robust for strict- and nearly-affine-related scenes, while providing a useful approximation for the overlap regions in scenes related by a projective homography or a more complex transformation, allowing for a set of efficient and accurate initial conditions for pixel-based registration.

We present an evolution equation-based binarization method for document images produced by cameras. Unlike the existing thresholding techniques, the idea behind our method is that a family of gradually binarized images is obtained by the solution of an evolution partial differential equation, starting with an original image. In our formulation, the evolution is controlled by a global force and a local force, both of which have opposite sign inside and outside the object of interests in the original image. A simple finite difference scheme with a significantly larger time step is used to solve the evolution equation numerically; the desired binarization is typically obtained after only one or two iterations. Experimental results on 122 camera document images show that our method yields good visual quality and OCR performance.