In this paper, we describe a novel approach to image sequence segmentation. In this approach, the presence of moving objects is first detected through background subtraction, i.e., the difference between the current frame and a dynamically updated background. Then, moving objects are extracted from the background subtraction image. Experimental results on surveillance image sequences demonstrated the efficacy of the proposed approach and its improvements over previous background subtraction techniques.

In previous work we developed a partitioning strategy called constrained framing, which was used to implement an error diffusion neural network for digital image halftoning. The partitioning approach made it possible to implement the original technique in a distributed, parallel fashion. The resulting halftoned images were visually indistinguishable from the non-partitioned approach. Ulichney's halftone metrics of radially averaged power spectra and anisotropy also did not show noticeable differences between the two techniques. Even an intensive application of the power spectra metric (averaging 1000 halftoned images instead of 10) revealed the need for some additional metrics of quality. Taking advantage of a priori knowledge of the partition scheme's geometry, we develop a protocol for evaluating the subtle differences between halftoned images produced using the constrained framing algorithm and the non-tiled halftone. An observed relationship between the new metric and input image grayscale magnitude suggests the existence of normative values which may be used to examine the performance of other halftone algorithms which are applied in a similar segmented fashion.

The characteristics of impulse bursts in radar images are analyzed and a model for this noise is proposed. The model also takes into consideration the multiplicative noise present in radar images. As a case study, soft morphological filters utilizing a training-based optimization scheme are used for the noise removal. Different approaches for the training are discussed. It is shown that the methods used can provide an effective removal of impulse bursts. At the same time the multiplicative noise in images is also suppressed together with god edge and detail preservation. Numerical simulation results as well as examples of real radar images are presented.

Based on the previous work, this paper presents a model- based approach to the representation and recognition of 2D closed, bounded, free-form curves and simply connected objects. The method involves a new expression of the objects based on complex variables and conformal mapping. In the model-based approaches, a object in a data set is recognized by comparing and matching its model with the known models of the objects that are stored in a database. To develop practical systems for object recognition, several problems must be solved: 1) how to get the model of a data set. We use a technique based on complex variables and conformal mapping to map an object to a circle. The coefficient vector of the polynomial mapping function is stored then in a database and represents the data set model. 2) when the object coordinates system changes, the coefficients vector of the polynomial mapping function also change. However, the coefficients change in a predictable and known way. The object recognition and matching is based on computing and comparing an object invariant entity. This invariant entity is a function of the coefficients and is independent of the coordinate systems. We show the effect of noise on object recognition and position/rotation/scaling estimation. Finally, some experimental results are presented to illustrate the power and robustness of the presented technique.

Hu and Zernike moments have been widely used in many shape recognition and object classification tasks. The Hu moments are the projections of an image I(x,y) on the basis functions formed by the monomials xPyq of non-orthogonal nature. Zernike moments are computed based on a set of orthogonal polynomials over the interior of the unit circle x²+y²<EQ1 in the image center using from the image pixels mapped into the unit circle and ignoring pixels falling outside the unit circle. Although comparative studies show that the Zernike polynomials have better noise performance in the context of character recognition and capable of coping both geometry as well as illumination invariance in the context of multispectral texture classification, there is no study conducted focusing on the application of Zernike moments to color image databases. In this paper, we study the performance of Zernike moments against the moments of Hu, in the context of RTSS (Rotation, translation, scaling)-invariant color scene classification, and determine which method behaves better in noisy visual information system. The method described here relies on the principal component analysis (PCA) and the discrete Karhunen-Loeve (KL) transformation for converting a vector- valued color image into a gray-level scale representation. For a selected set of images involving different models of vehicles, we demonstrate experimentally that the first two Zernike moments are highly immune to the sensory noise and also invariant to changes in the position, orientation, and size. They provide better classification power than the Hu moments in the presence of Gaussian noise.

Feature selection methods are useful to obtain an optimal set from a larger set thereby eliminating redundancy. In this paper, the popular methods of principal component analysis, Fisher discriminant analysis, and a genetic algorithm based approach are implemented for texture feature selection. The feature set is constituted by wavelet features. The selection processes are judged on using the classification rate of a particular classifier as a criterion. The Euclidean distance measure is used. The results show that the Fisher method performs better than principal component method. Also, the experiment concluded that the genetic algorithm improved the efficiency of all the methods except the Fisher method. Results of computation time are also presented.

An important aspect of long-term earth remote sensing missions, such as NASA's Earth Observing System, is characterization and monitoring of the sensor performance over the lifetime of the mission, which can be many years. Emphasis has traditionally been placed on the sensor radiometric response, but the spatial response is equally important for inter-sensor comparisons, accurate spatial pixel aggregation, and spectral signal unmixing. In this paper, I present a review of techniques for sensor spatial response measurement from operational imagery and contrast their relative advantages and disadvantages. Examples of spatial response and MTF measurement from airborne and satellite sensors are presented, and the sensitivity of different techniques to aliasing and noise and their requirements for target calibration are discussed.

This paper describes a method for assessing the information density and efficiency of hyperspectral imaging systems that have spectral bands of non-uniform width. The information density of the acquired signal is computed as a function of the hyperspectral system design, signal-to-noise ratio, and statistics of the scene radiance. The information efficiency is the ratio of the information density to the data density. The assessment can be used in system design, for example, to determine the number and size of the spectral bands. With this analysis, hyperspectral imaging systems can be tailored for scenes that are non-homogeneous with respect to spectral wavelength. If the scene spectral autocorrelation at each wavelength is different, then the information density at each wavelength is also different, suggesting that the spectral bands should have variable width. Two experiments illustrate the approach, one using a simple model for the scene radiance autocorrelation function and the other using the deterministic autocorrelation function of a hyperspectral image from NASA's Advanced Solid-state Array Spectroradiometer (ASAS). The design with non-uniform bandwidths yields greater information efficiency than an optimal design with uniform bandwidths.

Target detection in satellite images requires multilevel processing. One of the important levels of processing is image segmentation. In this, a gray level image is segmented into some limited number of classes. One of these segmentation techniques is clustering. Cluster analysis is the formal study of algorithms and methods for grouping. Most of the K-means class of methods start with some initial seed points and grow cluster around them. There exists two basic problems related to all seed based techniques - choice of appropriate initial seed points and single seed based techniques are not effective of more complex and elongated shape. Any elongated or non-convex cluster can be considered as the union of a few distinct hyper-spherical clusters. A multi-seed clustering method based on the border points proposed by Chaudhuri and Chaudhuri but, the border point extraction from remote sensing satellite imagery (RSSI) and noisy data is a difficult job. RSSI data has three basic characteristics - non-parametric, highly-overlapping and non-Gaussian. The proposed seed-point detection technique is statistical based and can solve the non-parametric as well as non-Gaussian problems. Also it focuses on the non-Gaussian field as the union of several Gaussian fields. Each seed point has a core set. After finding the seed points, an initial splitting technique (IST) based on the min-min-metric operation of the parameters mode and mean of each core set is proposed. Since RSSI data is highly overlapping, so IST algorithm can not cluster properly in the overlapping region. Another type of splitting technique is introduced in the overlapping regions. For finding the pair of overlapping clusters and overlapping regions minimal spanning tree based and set theoretic based mathematical formulation have been developed in this paper. Finally, the optimal splitting technique in the overlapping region based on maximizing the between class variance is introduced. This algorithm is tested on several real life data.

In this paper, we present the 2D Complementary Filter (CF) Banks, a signal processing technique applied to information extraction by edge detection. Based on the 2D multirate signal processing theory and complementary filters properties, the CF Banks allow to consider different types of sampling and filters. Procedures to design 2D nonseparable quincunx, rectangular and circular complementary filters are developed for an alias free decimation and interpolation. The CF bank is related with wavelet theory. Perfect reconstruction is achieved when the CF bank is implemented with quincunx sampling and filters. Although perfect reconstruction is not reached in other cases, the analysis and synthesis are performed without aliasing, corresponding to a high signal-to-noise ratio and visual quality. New structures were designed to get a better data compression. Implementation of CF Banks analysis/synthesis with images and application in information extraction by edge detection are shown.

Multiplex optical systems are designed to mix information channels on sensors. This paper considers the use of multiplex sensors to embed scene analysis in the optical layer of imaging systems. The goals of embedded analysis are to reduce the system data and processing load and to improve analytic accuracy by low-level filtering. We show specifically that filtering on the cross spectral density during detection allows the construction of sensors that integrate spatio-spectral patterns from specific volumes and that coded illumination patterns can structure the radiated cross spectrum to enable digital focusing on particular source characteristics.

Classical optical design techniques are oriented toward optimizing imaging system response at a single image plane. Recently, researchers have proposed to greatly extend the imaging system depth of field, by introducing large deformations of the optical wavefront, coupled with subsequent post-detection image restoration. In one case, a spatially separable cubic phase plate is placed at the pupil plane of an imaging system to create an extremely large effective depth of field. The price for this extended performance is noise amplification in the restored imagery relative to a perfectly focused image. In this paper we perform a series of numerical design studies based on information theoretic analyses to determine when a cubic phase system is preferable to a standard optical imaging system. The amount of optical path difference (OPD) associated with the cubic phase plate is directly related to the amount of achievable depth of field. A large OPD allows greater depth of field at the expense of greater noise in the restored image. The information theory approach allows the designer to study the effect of the cubic phase OPD for a given depth of field requirement.

Imaging systems have historically been associated with visual applications. Example applications include conventional photography, satellite imaging, and scientific visualization. Performance metrics relevant to these applications have traditionally been focused on visual quality and for such applications many conventional optical design metrics (e.g., worstcase spot size) can provide acceptable performance.[1] Recent imaging applications however, have broken with this tradition and are concerned with information-sensitive performance metrics (e.g., bit-error-rate). Examples of such imaging systems include those used in page-oriented optical storage, optical interconnects, and various forms of task-specific imaging (e.g., target recognition).[2][3] The ubiquitous nature ofdigital imaging modalities has made the need for fidelity-sensitive imaging system design even more apparent, with sensors and signal processing hardware k outpacing the perfonnance of low cost optical components. An information-optimized imaging system is one that takes as its primary goal the information throughput of the system. Notice that this goal can be very different from that of forming a "pretty picture." In this paper we limit our attention to binary-valued objects and within this framework we consider two cases. The first case will focus on diffraction-limited imaging systems whose performance is quantified via the average pixelwise mutual information. In section 2 we define the information-theoretic space-bandwidth-product (SBP) of such a system and demonstrate the existence of an optimal object configuration. The second case we consider is more realistic and includes the optical aberrations that characterize real world lenses. In this case we use the average bit-error-rate (BER) of the output image to quantify the overall imaging system performance. In section 3 we present the results of this case along with a comparison between some information-optimized designs and those based on more conventional design metrics.

The history of the spatial aspect of color perception is reviewed in order to lay a foundation for the discussion of retinex image processing. While retinex computations were originally conceived as a model for color constancy in human vision, the impact on local contrast and lightness is even more pronounced than the compensation for changes in the spectral distribution of scene illuminants. In the MSRCR, the goal of the computation is fidelity to the direct observation of scenes. The primary visual shortcoming of the recorded image is that dark zones such as shadow zones are perceived with much lower contrast and lightness than for the direct viewing of scenes. Extensive development and testing of the MSRCR leads us to form several hypotheses about imaging which appear to be basic and general in nature. These are that: 1) the linear representation of the image is not usually a good visual representation, 2) retinex image enhancements tend to approach a statistical ideal which suggests the existence of a canonical visual image, and 3) the mathematical form of the MSRCR suggests a deterministic definition of visual information which is the log of the spectral and spatial context ratios for any given image. These ideas imply that the imaging process should be thought of, not as a replication process whose goal is minimal distortion, but rather as a profound non-linear transformation process whose goal is a statistical ideal visual representation. These insights suggest new directions for practical advances in bringing higher levels of visual intelligence to the world of computing.

In this paper, we investigate the rate at which information is transferred from the scene to the digital processed image in imaging systems using nonlinear Local Normalization. While a formula for the Rate of Information has been used in many studies of linear systems with Gaussian (Normal) signals, a formula applicable to nonlinear/non-Gaussian systems has not been available until now. We discuss a new formula for the Rate of Information developed by the authors for systems with nonlinear Local Normalization processing. The Local Normalization algorithm, which is similar to the Retinex algorithm, is formulated and discussed. As the name implies, this algorithm forms a local average signal and normalizes the detail (high spatial frequency) scene signal by the local average. The scene detail information transferred to the locally normalized image is seen to be at least as high as that transferred to the acquired digital image. Thus, no information loss occurs in using Local Normalization. A case study of the new Rate of Information formula shows that designing the image-gathering system cut-off frequency near or slightly below the Nyquist frequency maximizes the scene detail information transferred to the locally normalized image for all the conditions considered.

This paper proposes a two-stage global motion estimation method using a novel quadtree block-based motion estimation technique and an active mesh model. In the first stage, motion parameters are estimated by fitting block-based motion vectors computed using a new efficient quadtree technique, that divides a frame into equilateral triangle blocks using the quad-tree structure. Arbitrary partition shapes are achieved by allowing 4-to-1, 3-to-1 and 2-1 merge/combine of sibling blocks having the same motion vector . In the second stage, the mesh is constructed using an adaptive triangulation procedure that places more triangles over areas with high motion content, these areas are estimated during the first stage. finally the motion compensation is achieved by using a novel algorithm that is carried by both the encoder and the decoder to determine the optimal triangulation of the resultant partitions followed by affine mapping at the encoder. Computer simulation results show that the proposed method gives better performance that the conventional ones in terms of the peak signal-to-noise ration (PSNR) and the compression ratio (CR).

This paper presents a rate-distortion (R-D) optimized 3D wavelet video coder by extending the concept of space- frequency quantization (SFQ) from 2D to 3D. A lifting-based 3D (1T+2D) wavelet transform is deployed to process one part of the sequence at a time continuously, thus eliminating the boundary effects over GOPs. After 3D SFQ, in which an R-D based spatial-temporal tree-pruning process is used in conjunction with uniform quantization of wavelet coefficients, we apply an efficient progressive 3D entropy coder to further compress the quantized coefficients. Even without motion estimation, the new 3D SFQ video coder outperforms MPEG-4 and other 3D wavelet-based video coders for most sequences at the same bit rate.

In this paper, we present a novel edge-based coding algorithm for image compression. The proposed coding scheme is the predictive version of the original algorithm, which we presented earlier in literature. In the original version, an image is block coded according to the level of visual activity of individual blocks, following a novel edge-oriented classification stage. Each block is then represented by a set of parameters associated with the pattern appearing inside the block. The use of these parameters at the receiver reduces the cost of reconstruction significantly. In the present study, we extend and improve the performance of the existing technique by exploiting the expected spatial redundancy across the neighboring blocks. Satisfactory coded images at competitive bit rate with other block-based coding techniques have been obtained.

This works deals with a new technique for large remote- sensing data compression. Contour mapping of two dimensional objects is of fundamental importance in remote sensing and computer vision applications. We present extensive algorithms applied to polygonized, simply- connected contours and reproduce desired shapes using an innovative data compression technique based on conformal mapping. In a previous work, through a conformal mapping process, we demonstrated the ability to 1) recognize shapes, and 2) concisely represent shape boundaries using a set of polynomial coefficients derived in the mapping process. In this work we illustrate how these previous results can be applied to data compression. In particular, in the approach outlined herein, a syntactic representation is formed for polygon shapes whose representation we desire to extract and reproduce compactly. Additionally, we present a problem of concavity in shape boundaries and a proposed solution in which polygons are divided into convex subsets and reconstructed accordingly. Each convex subset is then being processed in parallel, which lends to the extension of computational platform to parallel/distributed environment to improve the processing time. We show the potential of the proposed generalized technique in its ability to handle both polygonal, non-polygonal and mixed polygonal/non- polygonal object shapes.

The enhanced thematic mapper (plus)(ETM+) instrument on Landsat 7 shares the same basic design as the TM sensors on Landsats 4 and 5, with some significanti mprovements. In common are six multispetral bands with a 30-m ground projected instantaneous field of view (GIFOV). However, the thermal-IR (TIR) band now has a 60-m GIFOV, instead of 120- m. Also, a 15-m panchromatic band has been added. The artificial neural network (NN) image sharpening method described here uses data form the higher spatial resolution ETM+ bands to enhance (sharpen) the spatial resolution of the TIR imagery. It is based on an assumed correlation over multiple scales of resolution, between image edge contrast patterns in the TIR band and several other spectral bands. A multilayer, feedforward NN is trained to approximate TIR data at 60m, given degraded (from 30-m to 60-m) spatial resolution input from spectral bands 7,5, and 2. After training, the NN output for full-resolution input generates an approximation of a TIR image at 30-m resolution. Two methods are used to degrade the spatial resolution of the imagery used for NN training, and the corresponding sharpening results are compared. One degradation method uses a published sensor transfer function (TR) for Landsat 5 to simulate sensor coarser resolution imagery form higher resolution imagery. For comparison, the second degradation method is simply Gaussian low pass filtering and subsampling, wherein the Gaussian filter approximates the full width at half maximum amplitude characteristics of the TF-based spatial filter. Two fixed-size Nns (that is, number of weights and processing elements) were trained separately with the degraded resolution data, and the sharpening results compared. The comparison evaluates the relative influence of the degradation technique employed and whether or not it is desirable to incorporate a sensor TF model. Preliminary results indicate some improvements for the sensor model-based technique. Further evaluation using a higher resolution reference image and strict application of sensor model to data is recommended.

Situation when Point Spread Function (PSF) of the device is indefinite or it is roughly appreciated takes place almost always. In an optical wave range PSF (optics + atmosphere) is frequently indefinite, because of an indefinite conditions of scattering and turbulence of an atmosphere. The antenna pattern or PSF of radio telescopes cannot be measured. The near zone of the radio telescopes is so big that it does not allow to measure the antenna pattern or PSF by the ground based methods. The methods of evaluation of PSF on details of the images are considered in the repot. The received estimation of PSF is used for the solution of task (compensation of PSF distortions) super-resolution.

The concepts Point Spread Function (PSF) and Modulation Transfer Function (MTF) are base din the linear and local- linear methods of super-resolution. The resolving function, noise factor and value of achieved resolution are introduced into consideration. It is possible to make the definitions RPSF and RMTF for the resolved images. Values of the achieved resolutions will be estimated by the ratio of integral from RMTF to integral from MTF. It is possible to make the definitions NPSF and NMTF for the nonlinear iterative methods and similarly to estimate values of the achieved resolution. The report is devoted to the comparative analysis of the methods of super-resolutions.

This paper addresses progressive wavelet image coding within the trellis-coded space-frequency quantization (TCSFQ) framework. A method similar to that in [2] is used to approximately invert TCSFQ when decoding at rates lower than the encoding rate. Our experiments show that the loss incurred for progressive transmission is within one dB in PSNR and that the progressive coding performance of TCSFQ is competitive with that of the celebrated SPIHT coder at all rates.

It is known that gainful interframe compression of magnetic resonance(MR) image set is quite difficult problem. Only few authors reported gain in performance of compressors like that comparing to separate compression of every MR image from the set (intraframe compression). Known reasons of such a situation are significant noise in MR images and presence of only low frequency correlations in images of the set. Recently we suggested new method of correlated image set compression based on Karhunen-Loeve(KL) transform and special EZW compression scheme with strong zerotrees(KLSEZW). KLSEZW algorithm showed good results in compression of video sequences with low and middle motion even without motion compensation. The paper presents successful application of the basic method and its modification to interframe MR image compression problem.

In this paper, we propose a 3D directional interpolation algorithm for brain magnetic resonance (MR) images using neural networks. Typically, brain images consist of a number of 2D images. Although the sequences of 2D images provide basic information on structure, abnormality and etc., further processing on the sequences provides far more information. In processing 3D images, interpolation operation is one of the most widely used operations. In most conventional interpolation algorithms in the 3D space, the interpolation operation is performed separately in each coordinate that is orthogonal to each other. However, since the shape of the brain is roughly a sphere, interpolation along three orthogonal coordinates may result in some distortion, particularly in the vicinity of the boundary. In order to address this problem, we propose a new 3D interpolation algorithm. In the proposed method, we first perform the interpolation along two orthogonal coordinates. In order to find the best interpolation in the remaining coordinate, we search various directions that are not orthogonal to the two orthogonal coordinates using a cost function. Then we use neural networks to determine the final direction for interpolation. Experiments with brain MR images show improved results.

Image signal processing depends on computation intensive programs, which include the repetition of sequences of operations coded as nested loops. An effective technique in increasing the computing performance of such applications is the design and use of Application Specific Integrated Circuits using loop transformation techniques, and in particular, multi-dimensional (MD) retiming. The MD-retiming method improves the instruction-level parallelism of uniform loops. While many have written about the multi-dimensional retiming technique, no results have been published on the possible limitations of its application. This paper presents an analysis of that technique and its constraints when applied to nested loops with known index bounds, such as those found in two and three dimensional image processing.

This work deals with a new distributed face recognition technique based on transformation invariant conformal mapping. Face recognition using digital images is constrained by several factors like rotation (in and out of plane), scaling, and is usually operated under strict lighting conditions. In a previous work, through a conformal mapping process, we demonstrated the ability to 1) recognize shapes, and 2) concisely represent shape boundaries using a set of polynomial coefficients derived in the mapping process. In this work we illustrate how these previous results can be applied to face recognition. Additionally, in the approach outlined herein, a syntactic representation is formed for polygonal and non-polygonal shapes representing a given face features whose representation we desire to extract and reproduce compactly. In particular, in this paper we focus on the face outline, left eyebrow, right eyebrow, left eye, right eye, nose, curve form the lower nose to upper lips, and lips as face features. Some of these face features are grouped together and then processed in parallel in a distributed network of workstations via a Message Passing Interface running over TCP/IP. In particular, we show that a master-salve paradigm is used here to implement the proposed parallel and distributed algorithm and is base don a distributed-memory approach. Test were performed with 1,2, and three workstations. The algorithm assumes no constraints in the lighting, and the size of the window that has the face. This work deals with face representation and recognition and face segmentation is not the subject of this paper. We finally show the potential of the proposed generalized technique in its ability to handle both polygonal, non- polygonal and mixed polygonal/non-polygonal object shapes. We also show that the proposed algorithm achieves high recognition rates for rotations in the plane, translation, and scaling and is robust in noisy environments.

In this paper, we introduce a technique for handwritten words and sentences representation and recognition. The proposed method is based on complex variables and conformal mapping methodology. In particular, in a previous work, through a complex variable methodology and conformal mapping process, we demonstrated the ability to recognized shapes and concisely represent shape boundaries using a set of polynomial coefficients derived in the mapping process. In this work we illustrate how these previous results can be applied to hand-written words and sentences. We show that the words/sentences classification techniques used are adapted to the feature-coefficients selected and are based on feature-coefficients similarities in combination with the minimum distance classifier. We use as measures the Euclidean distance as well as the covariance matrix eigen- values distance. Finally, experimental results of handwritten words and sentences are shown to show the power, versatility and robustness of the proposed technique.

This paper presents a general spectral model for target detection in the visible spectrum and a cluttered environment. Having the model, the paper derives the optimal spectral filter to be attached to the camera's aperture plane to improve the detection abilities. Such filter may also be implemented by changing the integration periods of the R,G and the B channels of the CCD camera. Simulations demonstrate the abilities of the suggested approach.

In this paper, we present implementation of the robust RM- estimators with different influence functions such as the cut median (skipped median) function and Hampel function. We obtained that use of these functions in the RM algorithms demonstrated better robustness in comparison with the simplest cut median function. Applications of these functions in filtering procedures provide the preservation of fine details, impulsive noise removal and suppression of the multiplicative noise. The implementation of the cut median and Hampel functions in the RM-KNN filter has shown that its use is a good tool for preservation of fine details and suppression of noise by means of use DSP TMS320C6701. The deterministic and statistical properties of the designed filters have been investigated and shown their effectiveness. The optimal values for parameters of these filters for different noise mixture are presented in this paper. Finally, DSP implementation has demonstrated that in the case of use the simplest cut median the time of processing is less than in the case of applications the cut median and Hampel functions, but nosie suppression is better when cut median or Hampel functions were applied.

The studies made by using the domain of graphic images allowed creating facilities of the artificial intelligence for letters, letter combinations etc. for various graphics and prints. The work proposes a system of recognition and identification of symbols of the Arabic graphics, which has its own specificity as compared to Latin and Cyrillic ones. The starting stage of the recognition and the identification is coding with further entry of information into a computer. Here the problem of entry is one of the essentials. For entry of a large volume of information in the unit of time a scanner is usually employed. Along with the scanner the authors suggest their elaboration of technical facilities for effective input and coding of the information. For refinement of symbols not identified from the scanner mostly for a small bulk of information the developed coding devices are used directly in the process of writing. The functional design of the software is elaborated on the basis of the heuristic model of the creative activity of a researcher and experts in the description and estimation of states of the weakly formalizable systems on the strength of the methods of identification and of selection of geometric features.

Development of areas of automated systems of management of technological processes and systems of local automation requires the resolving of a set of questions on identification of production operations, working out industrial methods of measuring and control. Program-technical complex containing the systems of artificial vision, integrating device and dynamic expert systems of ready-product quality control in the example of decorative tile are examined at this work. The problem of identification of image can not be fully formalized and solved with the usage of strict algorithmic procedures and mathematical methods. Due to the mentioned fact, the development of intellectual programming methods- expert systems of image identification should provide effectiveness of mathematical methods of processing and heuristic programming with the expert knowledge of characteristics in analyzed systems. Implementation of the proposed complex, spares the specialist from routine job, allows timely spotting of technological process, solves the problem of sorting of ceramic materials in real time frame. In the meantime, the implementation of the system in dialog mode gives suggestions and recommendations.

In this paper, we proposed a stereo object tracking system that can control the convergence angle and pan/tilt of cameras by using optical binary phase extraction joint transform correlator (BPEJTC) and can extract the tracking object from a complex background and foreground noises by using the block matching-based window mask. It is used to perceive and extract the tracking object from the foreground and complex background by using window mask of the block matching-based SAD(sum of absolute difference), and by using the optical BPEJTC of the phase type, which has improved the correlating properties of the conventional optical JTC(joint transform correlator) with the adaptive object tracking ability, the position values of moving target on the left and right images can be calculated. And, we can be controlling the convergence angle and pan/tilt of cameras by using this values. Therefore, real time stereo object tracking system, which could adapt to the changes in surrounding, can be implemented. From the experimental results, the proposed stereo tracking system is found to track the object adaptively under the complex circumstances and changing background noises and the possibility of real-time implementation of the proposed system by using the optical system is also suggested.

AIM OF THE PRESENT WORK Copper (I) salts are unstable in many orgnic solvents. Previous investigations of copper(I) salts are limited only to acetonitrile and its binary mixtures with some organic solvents. Benzonitrile like acetonitrile has also -C -N group attached to the benzene ring system. This solvent like acetonitrile also stabilize copper (I) salts strongly. Moreover, the solubility of copper(I) salts in benzonitrile and its binary mixtures with other organic solvents is much higher as compared to that in acetonitrile. In spite of very favourable properties of benzonitrile, studies of copper salts have never been made in benzonitrile and its binaru mixtures with other organic solvents. The binary mixtures of benzonitrile with acetonitrile and pyridine has a wide range of dielectric constant and viscosity. these solvent systems are very interesting solvent systems for the study of copper(I) salts. Cu+ and Na+ ions have almost some ionic radii (Cu+ = 0.90 A and Na+ = 0.95 A). These ions are expected to have some type of solvation behaviour in many solvents. It has been found that in nitrile solvents and pyridine Cu+ is solvated by a special type of interaction while Na+ is solvated by the normal electrostatic ion-dipole interaction.