Detection of curves with special shapes has been put on great interest in the fields of image processing and recognition. Some commonly used algorithms such as Hough Transform and Generalized Radon Transform are global search methods. When the number of parameters increases, their efficiencies decrease rapidly because of the expansion of parameter space. To solve this problem, a new method based on Genetic Algorithm is presented which combines a local search procedure to improve its performance. Experimental results show that the proposed method improves search efficiency greatly.

In this paper, a new video description method based on finite element method is proposed. In the method, a finite element mesh is utilized to approximate an original intraframe image, this mesh is also utilized to describe motion between consecutive interframes. By applying the method presented in this paper, drawbacks in the conventional block matching method can be alleviated, and new video coding scheme can be developed.

In this paper an integrated vision system for autonomous land vehicle is described. The vision system includes 2D vision, 3D vision, and information fusion. The task of 2D vision is to provide physical information and 3D vision is to detect obstacles in the surrounding. The 2D and 3D information fusion can generate a feasible region provided for vehicle.

`Rainbow Range Finder' is a method for rapidly acquiring 3D information based on spectral analysis. It uses a special light with continuous spectrum to project across the objects, and its image will present the regular change of colors. One color forms a line in the color image and responds to a light plane structured by a wave band in the spectrum. When all the light planes are calibrated and the camera model is known, we can calculated 3D coordinates of all image points in the scene. This paper mainly discusses the light plane calibration and color classification techniques for implementation of these methods.

A fast 3D reconstruction of space curve based on quadric segment and matching in trinocular vision is proposed in this paper. First by using the perspective invariance of zero curvature and corner points in curves, curves can be segmented into different quadric curves. Then match these segments among three images and reconstruct them respectively. Finally combine all reconstructed segments together to form continuous curves. The experiments illustrated that the reconstruction speed is higher than those which take points or lines as primitives and duality and a uniqueness solution can be ensured.

A 3D shape retrieval by orthographical reconstruction based on an one-shot active system is presented. First its basic ideas and mathematical analysis are presented. Then a new method to determine crucial system parameter is proposed. Finally a computer synthesized experiment is conducted to evaluate our method. The experiment result is promising.

A wide field CCD camera has been developed and mounted on a small size pilotless airplane for photogrammetric test. Since the focal length, relative aperture and swath of the camera are closely related to the size of the CCD pixel, and the resolution and signal to noise ratio of the camera are determined by the CCD characteristics. Great care has been paid on the selection of CCD at the beginning of the system design, and great effort has been made to decrease the output noise of the camera, which depends on CCD and circuit noise. As the pilotless airplane is small in size and light in weight, the effect of wind stream and direction to the plane attitude should be taken into account. This paper presents the design of the pilotless airborne wide field CCD camera. Two main parameters, namely the resolution and signal to noise ratio of the camera are discussed, and the effect of attitude variation of the plane to the image taken is analyzed, and measures taken to overcome the effect and flight test results are presented.

We propose a 3D image display system which can present real scenes with realistic motion parallax. In the sensing system, a scene is observed by using a camera matrix. An excellent stereo algorithm SEA which utilizes 3 X 3 image matrix recovers the depth information of the scene with the density and the sharpness required for high quality image generation. In the display system, following the viewing position of the observer, 2D images with proper motion parallax are generated and presented. A prototype system has been developed and evaluation of the system is examined.

In this paper, real terrain is regarded as a hierarchical structure, by simulating the two main processes in earth evolution: construction movement and erosion effect, we integrate findings of classical geometry, fractal geometry, wavelet theory and image processing technique to propose a realistic and feasible hierarchical method for 3D terrain generation. Given some specific terrain parameters, e.g. standard deviation of terrain elevation, relative length, mountain trend, terrain coarseness, `real' 3D terrain can be automatically generated on computer by simulating the process of terrain evolution.

A novel 3D-terrain construction algorithm is represented in this paper. Traditional construction algorithms based on terrain self-similitude often ignore its structural features. This algorithm not only guarantees the integral and local self-similitude of constructed terrain, but also describes its structural features realistically. Iterated function system is used to construct terrain structural lines first. Then heights are got and adjusted by washing- out algorithm. Structural features of constructed 3D-terrain with this algorithm is conforming to the natural scenes, so it is more similar to natural scenes.

3D route planning is a challenge problem due to its huge computational cost burden. There are many algorithms about route planning such as A* and dynamic programming. But these algorithms are planned only from 2D plus 1D, instead of 3D directly. Generally, they first search a 2D optimal route in the ground, then plan their altitudes in a vertical profile determined by 2D route planning. In this paper a 3D route planning method based on genetic algorithm is proposed to select a 3D route directly. Through the experiments with a natural 3D terrain map, it is concluded that the route planned by this algorithm is satisfactory.

Threat source such as radar, is one of the most important factors which must be considered in route planning. Some threat sources generate so large threat area that the whole planning field is covered by it. In these areas, some threat areas can be observed by threat source, called visible area while others can not be observed due to the overlapping of terrain, called invisible area. In order to lessen the route planning cost, invisible area should be utilized as more as possible. In this paper a fast route planning approach which uses invisible area and the high maneuverability of aircraft is presented. After the comparison of theoretical analysis with experimental results, it is concluded that our approach needs much less planning time than the traditional algorithm.

This paper presents a novel method to analyze the matching suitability of reference map for mission planning of unmanned aircraft navigated by scene matching. The method is based on the relationship between matching probability of reference map and that of its points. First it estimates the matching probabilities of every point of reference map, then the matching probability of reference map is calculated by summing up these probabilities with weight of probabilities of navigation position distribution. Finally the experimental result is shown to demonstrate the validity of the method.

In this paper, a characteristic function is defined and used to quantitatively characterize exponential stability of nonlinear continuous neural network. By utilizing the function, we address many important aspects of network, including global and local exponential stability; the estimates of the domain of attraction of stable equilibrium point; the estimates of convergent rate of the network trajectories. A sufficient and necessary condition for network to be locally exponentially stable is obtained. Our method is simple and practical, and our results generalize those in 1-3.

There are many approaches about 3D route planning such as A*, potential field, dynamic programming. All of them plan the route only in the way of space route planning. As well known, the shape of 3D terrain is one of intricate features in nature. Is there a relationship between flight route and 3D terrain? In this paper, we concentrately reveal this relationship and attempt to apply it to route planning, presenting a new approach of 3D route planning. Through the experiments in a natural 3D terrain map, it is concluded that this approach is efficient and very fast.

3D surface reconstruction is one of the hotspots in Computer Vision and Remote Sensing. Relative orientation is a important part of 3D surface reconstruction. The traditional method that finding the parameters of relative oriented model is Newton Method, but it is highly sensitive to the initial values of parameters. To overcome this shortage, we give a method of using penalty function to find the relative oriented model in this paper.

This article discussed the computerized automatic evaluation of weld radiographs. The radiation projection method (RPM) was introduced, and a modified calculation method based on RPM was proposed. The actual experimental results demonstrated that this new method improved the ability of RPM to suppress the interference of noise.

In this paper, a robustness analysis of discrete autonomous multi-agent system by linear matrix inequalities is proposed, the core equations and their inferences are given. And corresponding concurrent schemes for analyzing multiprocessing system are discussed.

Virtual Reality Systems can construct virtual environment which provide an interactive walkthrough experience. Traditionally, walkthrough is performed by modeling and rendering 3D computer graphics in real-time. Despite the rapid advance of computer graphics technique, the rendering engine usually places a limit on scene complexity and rendering quality. This paper presents a approach which uses the real-world image or synthesized image to comprise a virtual environment. The real-world image or synthesized image can be recorded by camera, or synthesized by off-line multispectral image processing for Landsat TM (Thematic Mapper) Imagery and SPOT HRV imagery. They are digitally warped on-the-fly to simulate walking forward/backward, to left/right and 360-degree watching around. We have developed a system HVS (Hyper Video System) based on these principles. HVS improves upon QuickTime VR and Surround Video in the walking forward/backward.

The article analyzes the present volume measuring and mapping system used for coal yard, and presents a principle which connects phase-modulating laser ranging with high- accurate angle scanning, to fulfill an effective system of measuring and mapping the volume of heaps of coal. This system can complete shape-mapping and volume-measuring over 200 m X 50 m coal yard within 20 minute, and the accuracy up to 0.5%. Also, some detail working principle, actual measuring result and performance analysis of the system are discussed in this article. Due to having solved the prompt (< 1/500) and accurate (3 cm) ranging problem under the condition of low-reflection objects ((rho) < 0.01), this system can be widely used for 3D measuring and mapping on bulky objects.

Lighting simulation and image synthesis for outdoor scenes is intractable due to the geometric and lighting complexity of such environment. In this paper we present an efficient framework to the problem of image synthesis for outdoor scene by combining the geometry-based and image-based rendering. With respect to the linearity of incident spectral power distribution and apparent reflectance, the receptor responses of an image under varying sun position and weather conditions can be calculated to generate the desired image. Our method is proved to be available and feasible for fixed scene and camera geometry.

This paper describes a scalable architecture model that is suitable for multimedia image storage. In the architecture, the storage system is connected with networks so that direct data transfer between the clients and the server is available. The clients can receive data from the storage subsystem directly without going through the server. Thus, the bottleneck of the server doesn't exist. The paper demonstrates the detailed storage architecture. Because of the new architecture, file system and lower level storage operations will change correspondingly. Therefore, corresponding file system support for the architecture is discussed.

From a dynamical system's point of view, we choose several sets of simple and special mathematical functions to generate complex computer images. We give examples of various patterns with special symmetry in different convergence conditions and coloring techniques.

A prototype was designed to simulate spectral packinghouse measurements in order to simplify fruit and vegetable damage assessment. A computerized spectrometer is used together with lenses and an externally controlled illumination in order to have a remote sensing simulator. A laser is introduced between the spectrometer and the lenses in order to mark the zone where the measurement is being taken. This facilitates further correlation work and can assure that the physical and remote sensing measurements are taken in the same place. Tomato ripening and mango anthracnose spectral signatures are shown.

In this paper, we analyze and compare the performance, both in spatial and frequency domains, of different moments. Our analysis shows that Legendre and Gaussian-Hermite moments better separate image features based on different spatial modes and Gaussian-Hermite moments are less sensitive to noise. Analysis on spectral performance shows that Legendre and Gaussian-Hermite moments separate different frequency bands more effectively than others. It is also shown that Gaussian-Hermite moments give an approach to construct orthogonal features from wavelet analysis results. Experimental results are reported.

This paper describes a new jigsaw puzzle solver using chromatic information as well as geometric shape. Three new puzzle assembly algorithms are developed and experimental results on their performance are provided.

In the simulation of tracking a target such as a ship with infrared imager in the sea environment, how to generate dynamic images of the target and background is the key issue. The radiation of the scene includes the radiation of the target, sky, and seawater. All of these radiation can be obtained with the software--LOWTRAN7. A method of laser quantification is used to convert the radiance into gray- scale. The gray-scale of anywhere and anytime is computed beforehand and stored as a database. Establishing the 3D model of the target and background is necessary. The scene can be established through 3D transformation. We program with OpenGL in Windows NT operating system. The material of the objects is emitted material and its value is equal to the gray-scale. During the course of animation, the material and the position of the scene changed frequently. It is interactive.

The squint side-looking imaging mode SAR has a squint angle at the side-looking direction. This imaging mode can obtain multi-resolution images of a same scene with the variation of squint angle. But the imaging mode has to overcome three problems: (1) compensate the range walk and curve; (2) develop a mapping technique that works even when the size, shape, and orientation of a resolution cell varies; and (3) fuse the different resolution images to obtain the best imaging results and the most messages of the imaging terrain. The paper analysis and present an imaging schedule including imaging algorithm, geometry correction method and a different resolution images fusion method.

At first, the radar target scattering center model and MUSIC algorithm are analyzed. How to efficiently set the parameters of the MUSIC algorithm is given through a great deal of simulated radar data in experiment. After that, according to measured data of two kinds of plane target on fully polarized and high range resolution radar system, author mainly investigated particular utilization of MUSIC algorithm in radar imaging, 2D radar images are generated for two targets measured in compact range. In the end, we draw a conclusion about the relationship between radar target scattering properties and imaging results.

This paper studies on the multiresolution analysis and classification of the textured synthetic aperture radar (SAR) image using the wavelet transform (WT). The tree- structured WT algorithm is first employed to decompose a real-world textured SAR image, and to quantify different texture types in the image. Then, the pyramid-structured WT algorithm is applied in the multiresolution classification of the image. The classification result obtained demonstrates the advantage of the WT in the textured SAR image classification.

Pseudo-random code continuous wave (PRC_CW) radar, which is low probability of intercept and has good ability of target detection in clutter, is widely used in fields of target detection, tracking, guiding, detecting underground, and so on. Temporal distribution function of land clutter was always studied and the backscatter from land was supposed spatially homogeneous. In fact, the backscatter from various resolution cells is obviously different, especially under complicated land background, since the number of clutter sources is too great. So it is very important for spatial correlation characteristic analysis of land clutter. In this paper, spatial correlation characteristic of land clutter for PRC_CW radar at X-band was analyzed. The time spread of radar echo exists while the radar works at low grazing angle. To PRC_CW radar, the time spread of radar echo results in that the receiving signal amplitude at a certain time is the sum of many spread codes. In this paper, spatial amplitude distribution and spatial correlation function of some typical land clutter under the condition of single code were firstly discussed. Then, we point out the problem of time spread of echo signal at low grazing angle and proposed a superposition method to simulate real clutter echo of PRC_CW radar. Finally we analyzed the spatial correlation characteristics of some typical land clutter.

In this paper, it is firstly pointed out that the raining, as an object in the air, will cause the time spread of radar echo signal. For the pseudo-random code continuous wave (PRC_CW) radar, working with transmitting and receiving continuous EM waves, the echo signal at a specified moment will be the superposition of some spread echo signals in a period of time before this specified moment. A superposition algorithm was proposed to calculate the PRC_CW radar signal. Based on the former previous results on the characteristics of rain clutter, the amplitude fluctuation and the spectrum of the echo signal of the PRC_CW radar are investigated with simulation method.

The Doppler frequency rate is an important parameter in synthetic aperture radar imaging, which is used to perform the azimuth compression. The Doppler frequency rate can be estimated through the measurement of the sensor trajectory and attitude or through the analysis of the characteristics of the target echoes. The latter method will generally give a high accuracy. Currently, there are quite a few techniques for estimating the Doppler frequency rate, with some having a high accuracy but requiring a large amount of computation, and others requiring a smaller amount of computation but having a lower accuracy. Based on the concept of minimum entropy, a new technique for estimating the Doppler frequency rate is proposed which not only gives a high accuracy but also requires a small amount of computation.

In this paper, the principle of how to make us of the inter- intro FM waveform to achieve a high range resolution has been analyzed and the formulas for a point target and an extended target have been derived. On the basis of the derived formulas, the two ways to process radar echoes for a high range resolution image have been typically discussed.

The technique of millimeter monopulse radar 3D imaging of maneuvering targets is presented, and some key techniques to obtain high quality image are discussed. The efficient of the proposed method is confirmed by simulation results.

Range-Instantaneous-Doppler Algorithm (RIDA) should be used in ISAR imaging when the target is maneuvering. With the assumption that Doppler shift of scatterers is low-order polynomial, an adaptive LFM `CLEAN' technique to estimate the instantaneous frequency is proposed, where scatterer Migration Through Resolution Cell is also considered. The final RID images of the real maneuvering aircraft are obtained, which confirms the effectiveness of the proposed technique.

2D phase unwrapping is the key technique in SAR interferometry. This paper discusses a phase unwrapping algorithm based on fringe line detection. Improvements are presented in order to use this method in areas with intensely fluctuated fringe lines. Results on X-SAR interferometric data are obtained. They confirm the excellent performance of the procedure.

A new numerical model to predict the IR radiative temperature of the ground with different mulch is presented. Based on the energy and water balance principle, the mulched and unmulched ground surface boundary conditions for both heat and water flow are described. The temperature of mulched and unmulched soil region and mulch surface are determined to predict the surface radiative temperature of the ground background. The inputs required for the numerical simulations are weather data, soil thermal and hydraulic properties, and mulch data. Numerical experiments are performed to examine the effect of soil type, mulch kind, and weather conditions on the spatial variations in radiative temperature. 24-hours continuous simulations for each combination of different soil type, mulch kind and mulch width show that the soil type, mulch kind and width, and weather conditions can evidently affect the radiative feature of the ground surface.

128 X 128 diffractive microlens arrays have been designed by considering the independent optical and processing parameters for 3 - 5 micrometers wavelength with a microlens size of 100 micrometers . The lens F number and array pitch are 1.5 and 100 micrometers , respectively. The diffractive microlens arrays have been fabricated on the surface of Si substrates by successive photolithography and Ar⁺ ion-beam-etching technique. The practical processes and fabrication method are discussed. The optical characteristics and measurements of the diffractive microlens arrays are presented.

Acquiring 3D color model of human head is desired in many applications. In this paper, we introduce a scheme to obtain 3D color information of human head from image sequence in 3D laser color scanner. Structured light technology is employed to measure depth. We study the relationship among the object's images in different position. Synthesizing these information, we can obtain the shape of hair area from contour image. True color information of sample points can be acquired from the specified image in image sequence. The result of experiment is satisfactory.

Basing on the analysis and simulation of a-Si TFT/PIN coupled unit, we obtained optimized design for structure parameters of detecting unit and material parameters suitable for detecting spectrum. In order to satisfy the low-temperature process of a-Si TFT and a-Si PIN, polyimide (PI) film has been used as the insulator in the TFT/PIN array. We have investigated process compatibility between a- Si TFT and a-Si PIN preparation and fine pattern for low- temperature PI film. As a result, we obtained 2D a-Si TFT/PIN image sensor (6 X 12 pixels) which has signal readout and sensor function. We tested and analyzed its static characteristic. The results indicate that the characteristics of PIN photodiode and TFT switch are better, and the performance of illumination response of the 2D image sensor is good. We propose to apply the image sensor into neural network system in order to decrease cross-coupling between pixels and increase parallel processing rate.

Spatial resolution improvement of digital images has significant applications in remote sensing and computer vision. In this paper, a general iterative method is proposed to improve the spatial resolution from low- resolution images in spatial domain. The general method of interpolation and simulated sampling is formed based on the iterative methods for signal reconstruction from nonuniform sampling and the methods of projection onto convex sets by defining convex sets of the sampled images and projection onto the sets, and using a general parallel projection method to find the common points of the sets. The method can be applied to multiframe images with different spatial resolution, various image radiance, relative geometric distortion, additive random noise, and some other general imaging style. Some experiments with resolution test pattern and multiangular remote sensing images performed the convergence and the effectiveness of the algorithms.

The key motivation for visual interpretation of hand gestures is to introduce this natural and intuitive communication mode to human computer interaction. Visual interpretation of hand gesture had typically made use of high-level parametric models representing the body parts such as arms, figures, palms etc. and their connections to each other. Such 3D model-based interpretation has been successful in some case; however, heavy computation makes this approach a very difficult task. By use of variable- order parameterized models of image motion and robust dominant motion regression, in this paper, we propose a motion-based segmentation scheme to feature extraction of hand gestures. The proposed scheme can directly estimate image motion of an object in two unsegmented images and obtain fine segmentation of that object from background at the same time. Based on inter-frame image motion parameters and the fine segmentation, we can construct various motion features, shape features, or their combinations for the purpose of hand gesture interpretation. With these features 12 kind of hand gestures used in our experiment can be reliably interpreted.

Optimization-based derivative-free line search methods are presented to locate all the corners of 2D polygon images for pattern recognition and image matching purposes. These optimization-based methods are also used to approximate a 2D non-polygon object by a polygon to desired accuracy. Experimental results are presented to compare these methods.

Imaging Spectrometer is a new remote sensing tool. Unlike conventional remote sensing, the Imaging Spectrometer can provide data in narrow, contiguous spectral bands across the solar reflected portion of the electromagnetic spectrum 0.4 um to 2.5 um. Appearance of the Imaging Spectrometer are providing opportunities for acquiring ever-increasing volumes of information as well as making more types of geological information potentially derivable. In order to capture the full range of geological information contained in huge volumes of sea-like data effectively, we have to develop new analytical tool for processing of hyperspectral remote sensing data. In this paper, authors generalized the basic theory and method of hyperspectral data processing, and pointed out the advantages and deficiencies of the various theoretical models. In addition, a practical experiment of data processing was conducted to perform the analytical processing technology developed in the paper. A error analysis was reported, some interesting suggestions were given.

In this paper, we present the output probability density functions of generalized open-closing (GOC) and generalized close-opening (GCO) filters under the different input distributions such as uniform, Gaussian and biexponential distributions, and calculate their digital features (expectations and variances). Then, we apply these filters to restore an image corrupted by impulsive noise and further test their efficiency in noise-suppressing and detail- preserving characteristics. The simulation results show that the GOC and GCO filter have good performances.

In this paper, a new method for extracting ellipses from an image is presented. It includes two steps, finding the candidate regions and rectifying the detected regions' position, estimating the ellipses' parameters. It's calculation amount is less than many other methods, and it is very immune to the noise. Specially, it has strong ability to extract multi incomplete and concentric ellipses from an image.

VQ (Vector Quantization) reduce the bit rate by exploiting the correlation in the data. To improve the performance of a compression algorithm based on VQ, this paper introduced a more efficient scanning method, i.e. Peanoscanning, which maintains better correlation in 2D data than that of raster scan, and then a hierarchical VQ based on the characteristics of image data is presented, at last we reduced the blocking effect by a smoothing algorithm.

In order to achieve high compression ratios while maintaining high visual quality, this paper propose a new hybrid zerotree and DPCM based wavelet coding scheme for still image. We further exploit the dependencies between neighboring pixels in the same subband and the dependencies between descendants and ancestors in this paper, and we apply an explicit quantizer to each coefficient, then we transfer and encode data by bitplane progressively to maintain an embedded characteristic.

In this paper, we propose an translation- and scale- invariant adaptive wavelet transform, and according to this adaptive wavelet decomposition, we construct a stable multiresolution feature vector of images which is translation-invariant and scale-invariant. At end, we give some experiment results to demonstrate the stability of this feature vector under translation and scale transform.

Discrete-Amplitude Multiresolution Analysis (DAM) is a new kind of multiresolution analysis that uses signal quantization resolution as its scale. In this paper, a new efficient image coding method that utilizes the DAM's 2 bits recomposing theory of arbitrary continuous signal is realized. The problem of how to choose the coding direction, which is important for the 2D image data while not exists in the 1D data processing, is emphasized. The problems appeared with the application of DAM on image data are also analyzed. The experimental results proved that DAM image coding is a simple, easy-approaching and efficient coding method and it should have more potential application in image processing.

The application of mathematical morphology to edge enhancement of images is studied in this paper, and a morphological edge enhancing operator and its anti-noises counterpart are also presented. This morphological operator enhances the gradients of edges while attenuates noises and preserves edges. It is showed by computer simulations that the gradient of edges is enhanced by the morphological edge enhancing operator and, then, the structure and feature of images are strengthened and edges can be easily extracted.

The typed association of objects for image analysis via structured quotient set is presented in this paper. From a landscape point of view, the elements of spatial structure are organized into some distinct patches, which are the smallest homogeneous units of landscape at the spatial scale we can see. The main advantage of such patches is that it gives us the typed association of objects. In order to remark the relationship between the components of an image corresponding to the patch, the notion of diffusion- concentration is introduced, whose practical meaning in the patch is to join a pixel to one another and determine the weights of the links. Another research goal is to explore the properties of variogram function, from which the weights of neurons for growing collection of the typed association of objects by quotient set can be derived.

This paper presents a new algorithm for the fast hough transform. Compared with others, this algorithm not only has a faster speed but also has advantages such as avoiding erroneous solutions, eliminating redundant points, detecting straight lines of different sizes, and selecting threshold automatically etc. We will discuss the performance of this algorithm and will also show its efficiency with examples.

A new segmentation algorithm of forward-looking infrared (FLIR) images is presented, which first uses median subtraction filter to enhance targets and suppress backgrounds, then uses MBS algorithm to perform segmentation. This algorithm can obtain a precise and accurate segmentation of a target from low contrast FLIR images in complex background. Experimental results compared with MBS algorithm are given. We find that the proposed algorithm shows much better segmentation performance than MBS algorithm in complex background.

Using the 3D scanner, a large quantity of 3D data set of the scanned object is generally acquired. It isn't very convenient to transform and to process such a large data set. So compression is necessary. In this paper the 3D data compressing of the human head surface is considered. A 2D curve compressing algorithm is extended to 3D space and is improved by adopting an adaptive threshold method. Experiments show that both the compressing rate and the reconstructing effect are satisfactory.

A sample of radio sources with ultraluminous infrared luminosity has been observed by using the Australia Compact Array at multiple wavelengths. The parameters, such as position, flux, spectral index and size, etc. of the radio sources have been obtained. The paper presents our data reduction of the aperture synthesis telescope and mapping method.

We propose a new Gibbs sampling algorithm, the soft- criterion acceptance algorithm, and use it for the texture synthesis. The new algorithm combines the advantages of the ICM algorithm in computations and of the algorithm of simulated annealing (SA) in global convergence. As a result, it is computationally efficient in comparison with the Gibbs sampler by S. Geman and D. Geman. The key idea is that the difference of the maximum and minimum of the energy functions is used to construct a soft criterion for updating each pixel value in a probabilistic acceptance fashion that is similar to the SA. The algorithm is verified by computational experiments.

A new adaptive algorithm for image zooming with edge preservation is proposed. The 4-point interpolation scheme is introduced to resample the original low-resolution image. By using the wavelet transform and inverse wavelet transform, edges in original image are extracted and restored to the final high-resolution image automatically.

We present a set of hierarchical expression rules based on which facial expression analysis in facial images can be carried out. The proposed expression rules describe an expression as a combination of facial actions, and a facial action as a combination of FFP movements. The FFP movements are specified both in direction range and in magnitude range. As the hierarchy of the expression rules indicates, facial expression analysis boils down to the FFP tracking. We adopt a model-based approach to accomplish this task. Possible applications of expression analysis are presented, which include expression recognition, facial image retrieval, and facial expression agent. Backgrounds of these applications are introduced and related key technologies are discussed.

Image morphing method, which is based on the pixels positions and color interpolation between two images, has been widely used in image interpolation. Existing morphing methods can not keep 3D shape, even though 2D shape. In this paper, we put forward a new morphing method. The interpolation image gained by our method is correspondent to certain viewpoint. The method need not predetermine the camera matrix. And it is of simple calculation and easy realization. Our method is practical by the experiments: the interpolation images gained by our method exactly represent the transformation of the view point from the resource image to the destination image.

Using JPEG standard, color still images can be compressed in high ratios while good quality can be guaranteed. The framework of the JPEG is specified, but the recommended quantization table and the Huffman table can be pruned according to the image's characteristic. In this paper, we put forward an adaptive-quantization approach which is compatible for the baseline JPEG. In our approach, we statistically calculate the image's average probability distributivity, which used as the threshold for the quantization. The quanti-table recommended by JPEG can then be justified with the characteristic of images and the average code-rate is noticeably dropped.

This paper describes a new system for extracting bibliography regions from the color images of a book cover. Following the preprocessing, there are color space segmentation and text region extraction processes. By comparing the text regions on front cover to those on spine, all text regions are classified into author region, title region, publisher region or other information region furthermore, and a bibliography image is obtained as a result.

In this paper, a new threshold method based on morphological operations and the fourth central moment criterion is proposed in order to segment gray-scale images into bilevel images. The method consists of four steps processing. Experimental tests on text analysis under various non- uniform background illumination show very good performance for extracting words from the background.

The images processed by our brain represent our window into the world. For some animals this window is derived from a single eye, for others, including humans, two eyes provide stereo imagery, for others like the black widow spider several eyes are used (8 eyes), and some insects like the common housefly utilize thousands of eyes (ommatidia). Still other animals like the bat and dolphin have eyes for regular vision, but employ acoustic sonar vision for seeing where their regular eyes don't work such as in pitch black caves or turbid water. Of course, other animals have adapted to dark environments by bringing along their own lighting such as the firefly and several creates from the depths of the ocean floor. Animal vision is truly varied and has developed over millennia in many remarkable ways. We have learned a lot about vision processes by studying these animal systems and can still learn even more.

Human face location is an important basis for automatic face recognition systems. In this paper we defined one kind of symmetry transform called Directional Symmetry Transform (DST), which can provide more information about objects' shape for object recognition besides of symmetry. When applied to object detection, it can bring us much convenience. One fast algorithm based on DST was proposed for face location from frontal face image with simple background. Experimental results proved its effectiveness.

An active origin-based affine matching method was proposed in this paper. The basic idea is, the origin which has two adjustable parameters on the affine coordinate frame and the affine coordinate frame are both active. So the affine transformation between the model affine coordinate frame and the image affine coordinate frame has two adjustable parameters which will be applied to affine matching. Theoretical analysis and experiments demonstrated the proposed method being better than the conventional method.

New approaches to the problem of matching method for dissimilar handwritten character images are presented in this paper. One of them is for the offline recognition of dissimilar two-stroke digits which are difficult to discriminate. Several measures are taken to obtain proper division and description of the two strokes to further discrimination. In the recognition stage we combined syntactic analysis with statistical calculation. The other approach is for verifying offline handwritten Chinese signatures. Local connection properties are fully employed and the weighted distances are calculated on the whole signature image. Experiments results show that using the proposed matching methods increased the recognition rate and improved the reliability of our OCR system and is also helpful for verification of handwritten Chinese signatures.

Approaches to the prototype information processing system of imaging seeker (PIPSIS) with an Infrared Focal Plane Array (IRFPA) sensor and custom image processing hardware are described. The prototype system consists of general digital signal processors and specific infrared image processing hardware. An intelligent infrared image processing software with multiple tracking algorithms of projection, label and correlation has been developed in consideration of the specific needs of the seeker system. In an effort to compensate for the photoresponse nonuniformity in the IRFPA sensor, an improved piecewise-linear nonuniformity correction procedure is investigated. According to the environmental characteristics of imaging seeker, analog signal pre-processing and image sub-regions segmenting are considered. Experimental results show that PIPSIS could provide maximum track range, more robust tracking, and better false-target discrimination.

The paper presents a new approach to match an image to a model described by straight lines. It utilizes both straight lines and multi-properties of the line hierarchically, and works efficiently under arbitrary translations, rotations and scale changes in noise conditions. It is practical for many applications.

In this paper, the image matching problem is discussed mainly in the light of its use on aerial pictures where matching robustness and realtime are to be considered in the first place. Three methods, 2D Hopfield N.N., Hausdorff distance measure and learning from examples with the laboratory results obtained are presented. Laboratory results show the robust performance of the methods. Also, a realtime system based on an array processor L64720 is presented.

The image registration is one of the key points in the integrative analysis of image data. In this paper, a full- automatic method for registration between SPOT and TM image is described. The results of the experiment show that the method is much better than traditional manual approach both in the efficiency and the accuracy.

The minimum description length (MDL) criterion is used to fit a facet model of a car to an image. The best fit is achieved when the difference image between the car and the background has the greatest compression. MDL overcomes the overfitting and parameter precision problems which hamper the more usual maximum likelihood method of model fitting. Some preliminary results are shown.

A purely automated technique is proposed for detection of `grain' objects on pseudo-homogeneous background and statistical estimation of their geometrical parameters. The objects are distinguished from background by average brightness and have some characteristic shape. Combination of discriminant and structural approaches to pattern recognition are involved. Original boundary tracking algorithm is used at segmentation stage.

A novel approach for feature extraction of radar targets is proposed in this paper. This approach explores canonical analysis on a matrix formed by the range profiles (RPs) of training targets in different aspect angles. A subspace is obtained in this analysis. Projection of a RP into this subspace forms canonical profile (CP). The CPs of a training target in different aspect angles are averaged into feature vector for this target in data base. Using CP of an unknown target as the feature vector for recognition, experimental results on simulated data are presented.

This paper is concerned with the research work about small target detection in heavy noise background infrared images. In the studied sea surface infrared images, ship objects are rather dim in dark sea background. There also exist scan line disturbance and clutter noises which increase the difficulties of exact detection. On one hand, dim objects must be detected from dark background. On the other hand, the small targets must be distinguished from clutters. Through the analysis of the targets and background features, we find the targets are more continuous then clutters and have sharper edges. Their sizes are larger too. These indicate that their frequency features are different from clutters. In frequency domain, the targets mainly lie in the low and middle frequency region compared with clutters. We apply low pass filters (LPF) to the images. The space sizes of LPF are carefully chosen according to the target sizes. To obtain good result, multi-level filters are considered. We subtract the filtered image from the original image, then use a contrast-based method to detect the objects. Our experiments show that the algorithm is excellent for target detection and robust to noises.

In this paper, a new small target detection approach which is based on the analysis of image background texture is presented, and Gaussian-Markov random field (GMRF) is chosen as texture model. Texture features can be obtained from the least square error estimation of GMRF parameters. Image is divided into non-overlapped blocks, and feature vector is obtained for each block. Using the normalized Euclidean distance of feature vectors as a measure of the difference of textures, target detection can be performed in a coarse to fine procedure according to the difference of features and approximation error.

Puny target detection is a common problem for image recognition. The main difficulty origins from that targets are putty and variable. This paper advances a detection algorithm called Flexible Receptive Field (FRF) that gives a new way to detect puny target. FRF is like as a `pupil' which can adjust itself size to acquire the maximum excitement intensity. Scanning a pending image with FRF, when a target falls into FRF, the excitement intensity will differ from that when there is not target. To reduce the overmuch computation due to flexibility, a fast algorithm with computation complexity O(n²) is advanced (similar to a normal filter operator). Experiments proved that FRF is a simple, fast and efficient method for puny target detection.

This paper presents a robust algorithm for detecting and tracking multiple targets in a long video sequence. What differentiate it from previous approaches are the complex environments, the unconstrained camera motion, as well as the comparatively short distance between the sensor and the targets. Firstly the background motion is estimated by a global LMedS algorithm, then on the compensated difference image a graph-like stochastic procedure is applied for tracking multiple moving objects. Real video experiments show its efficiency.

This paper addresses point-like target detection of dual- band infrared data. The adaptive spatial-temporal filtering algorithm is derived from the condition of local time- invariant Gaussian clutter interference. The simulation results with real world infrared images show that the algorithm has successfully suppressed the background clutter and the signal-to-clutter ratio is increased.

The character of the imagery for the stare infrared detector when the different characteristic targets are captured and tracked is discussed in this paper. The fuzzy control algorithm is presented. The fuzzy controller which is based on the above mentioned idea is studied. The practical system is studied experimentally. The results show that for the multi-input and output system the control method which is presented in this paper is provided with high control precision, the control performance of high speed response and extremely strong robustness.

In this paper, we propose a scheme of using superresolution range profiles (SRRP) as features to recognize radar targets and present an algorithm for SRRP's matching. The superresolution property of SRRP's relax their dependence on aspect and achieve a larger allowable aspect increment for the SRRP templates to fully characterize the target. The template library is reduced to a small size and the matching process is accelerated. Performance of our scheme is evaluated using a dataset of three scaled aircraft models. The experimental results show that the reduced template library can still achieve high recognition rates.

Efficient content-based retrieval of complex image is a challenging task since the queried object may appear in various scale, rotation and orientation with a wide variety of background colors and forms. We propose a novel method of detecting multi-colored object in image by content. Color component and spatial relationship are two important contents of image. With color feature based on hue histogram of HSV color space, we try to combine similar color pixels to a more homogeneous color component by adopting multilevel thresholding method based on the fourth central moment. And then, we use the extended color adjacency graph (ECAG) to describe a multi-colored object. ECAG consists of a set of nodes and two sets of edges. Each node represents a single color component of multi-colored object, while edges are divided into two classes: edges representing adjacency between similar color components and those between different color components. By investigating whether ECAG of object is the subgraph of ECAG of an image, we can detect whether the similar object exists in the image. Experimental results show this representation of object to be invariant to scale and rotation.

A new recognition algorithm using composite operator and direction curves to auto-detect the dreg defects in x-rays photo is proposed in this paper. This algorithm is simple with little computation, fast detection speed, good effect of real-time process, high accuracy of detection and good adaptability. Experiments evince this algorithm is a good feature detection method.

Personal identification is to associate a particular individual with an identity. It plays a critical role in our society in which questions related to identity individuals such as "Is this the person he or she claims to be?"

Images, photographed through a submarine periscope, are ambiguous because of submarine periscope light path energy dissipation and bad conditions. With wavelet and Mallat algorithm, we can remove image noise and enhance image edge. The processing efficient is not very good because the most optimum adjusting coefficients are not used in wavelet reconstruction. In this paper, we utilize the Orthogonal Experiment Design to optimize adjusting coefficients in enhancing edge and removing noises. Experiments show that the results are better.

Fractal compression is an attractive compression method with the ability of high compression ratio. But there are also some problems that has not been well solved yet. The wavelet-based fractal compression method proposed in this paper can not only make up its weakness, such as long encoding time, but also bring out its merits, such as high compression ratio.

In this paper, our purposes are image compression code by wavelet transformation, and develop a novel encoding algorithm structure of DPCM/WT on the basis of the DPCM/DCT. We have proposed a new encoding algorithm structure of DPCM/WT based on object driving and data flow driving, the novel algorithm in property is superior to DPCM/DCT in compression ratio, fidelity and real time processing.

Different type of wavelets has been constructed in order to be adapted for different applications. In this paper, we have got a new family of wavelets through self-correlation function of Daubechies wavelets and discussed their some properties and applications.

This paper proposed a method to segment the texture image based on multiscale analysis of wavelet transform. The features at different scales need study when the number of the textures is to be determined. The proposed method firstly transform the texture image with wavelet transform, then extract texture features on different scales and different frequency, and perform the coarse segmentation on those different channels of the same scale and the different scales. At last, the coarsely segmented results at same scale are incorporated together, followed by a inter-scale fusion procedure. Promising results have been achieved.

This paper analyzes the approximation performance of a special prefiltering algorithm for the orthogonal wavelet base. When the scaling factor is sufficiently large, we obtain a asymptotic formula of the approximation error of the prefiltering projection for the smooth signal. Also, we obtain a sharp upper bound of the approximation error of the prefiltering projection for the bandlimited signal. The results show that the convergent order of the prefiltering projection is the same as that of the orthogonal projection. With this one can determine the initial scaling factor and furthermore determine the corresponding initial scaling coefficients by the algorithm to satisfy the given error requirement. Finally, the experiments show that for the signals with normalized energy, the scaling function D₄ and J equals 2 may be generally selected to get the accurate initial scaling coefficients to some degree.

Wavelet, as a brand new tool in signal processing, has got broad recognition. Using wavelet transform, we can get octave divided frequency band with specific orientation which combines well with the properties of Human Visual System. In this paper, we discuss the classified vector quantization method for multiresolution represented image.

Recently there has been a considerable interest in the use of a somatosensory evoked potential (SEP) for monitoring the functional integrity of the spinal cord during surgery such as spinal scoliosis. This paper describes a monitoring system and signal processing algorithms, which consists of 50 Hz mains filtering and a wavelet signal analyzer. Our system allows fast detection of changes in SEP peak latency, amplitude and signal waveform, which are the main parameters of interest during intra-operative procedures.

This paper concentrates on two issues related to wavelet- based image coding. At first, an algebra method is proposed to select biorthogonal wavelets for higher compression ratio and better quality of the reconstructed image. Then, a general algorithm is given to convert a discrete biorthogonal wavelet transform into lifting steps, so that one version of the transform, which maps integers to integers, can support both the lossy and lossless coding to facilitate the region of interest based image compression.

This paper presents a fractal-based method for natural scene image segmentation. The main goal is to find artifact objects from complex natural scene. We propose a set of fractal measurements in order to acquire various aspects of roughness of each part of an image. The performance of the data fitting in the box-dimension estimation is analyzed and an improved algorithm is proposed. Experiments prove that the proposed approach is suitable for texture segmentation and artifact object finding in natural environment images.

Fractal coding of digital images offers many promising qualities. However the coding process suffers from the long search time in the Domain Block Pool. We present a novel idea for speeding up search and improving compressive ratio, based on the hypothesis that those correlative Range Blocks should have a common code. Experimental result on 512 X 512 Lenna image is 32.8 dB at 0.625 b/pixel and 29.76 dB at 0.214 b/pixel.

Since the encoding time for fractal encoding method is high and of importance, how to speed up the encoding is a key issue. In this paper, we propose a fast and efficient fractal coding algorithm, which is based on merged quadtree partitioning scheme. In this algorithm, the searching procedure in the finer level sufficiently uses the computing result of the coarser one. For the searches in finer levels, most of computation has been completed by searches in the coarser level, so the total search time reduces greatly. Besides, we use simple merged quadtree to reduce the number of transformation needed by encoding the whole image, so at the same PSNR, we also gain an increase of the compression ratio.

Fractal is everywhere in natura, therefore it can be used effectively in analyzing images of natural scenes. In this paper, multi-spectral image segmentation based on fractal characteristics is studied and its computational implementation is proposed. Our simulation shows that fractal signature performs as a good texture feature for segmentation of remote-sensing images.

The neural network method applied for the color image segmentation of the human's head image in the simple background is studied in this paper. The adopted network model is BP network. The image segmented by the region growing method is used as the training set and the BP algorithm is adopted to train the target and background image. Then segmenting testing image can be processed. Experimenting results show that the segmenting effect of this method is as good as the region growing method.

In data visualization, an efficient method for decreasing computation and storage is to extract feature region from massive data set and process (visualize, store or transmit) it solely. For this reason, in this paper, we will develop some algorithms for extracting region containing feature and for detecting edge points from 3D mesh data.

An unsupervised back-propagation neural network with texture discrimination is proposed for classification of multi- polarized SAR images. The first improvement of this method is to design an unsupervised training process for the back- propagation network. We use clustering methods to form initial clusters. Then a moving windows is used to pick the training sets automatically. By such preprocess the back- propagation network is unsupervised overall. Another improvement in this method is that, except for radiometric information, we integrate pixel texture into neural network. The parameters of pixel texture are calculated by gray level difference statistics in the preprocess phase and then are input to the nodes with gray-level of a pixel. The improved network is used to classify multi-polarized SAR data. We compare its result with that of BP network which use radiometric information only. The result shows that this method is effective and the classification performance is improved.

In this paper, a computer-aided diagnosis system for heart function based on artificial neural networks and fuzzy logic is introduced. Typical parameters reflecting heart function, provided by echocardiography, were used as input of neural networks and their corresponding heart functions as output. To obtain an analytic and discrimination model closer to brain, we combined fuzzy theory with neural network technology, and input parameters are fuzzily treated. During distinguishing morbid style, we used fuzzy interval, fuzzy number and its related possibility distribution concepts, and selected appropriate operator, and so get its corresponding membership, meanwhile membership was put out of interval of linguistic to consist with language expression. The network selected was BP, and back- propagation algorithm was used to train the network. After studying the result evaluated by expert, the neural network was used to appreciate 150 testees' heart function, of which 90.7% was consistent with experts' diagnosis.

A new model of feed-back neural network is presented in this paper and a simulative research work has been done on its dynamics behavior. The result show that equilibrium point and limit cycle may exist simultaneously when three neurons are in connection, additional chaotic phenomenon can occur with four neurons being in connection and the network characteristic has a close relationship with self-feedback intensity.

A mechanism turned a vision camera around a fixed carrot taking 100 images of it. A 3D reconstruction finite element algorithm reproduced the volume using finite area triangles and morphological operations to optimize memory utilization. Volume from several carrots were calculated and correlated against real volume achieving a 98% success rate. Three images 120 degrees apart were acquired and the main features extracted. A neural network system was trained using the features, increasing the measuring speed and obtaining together with a regression algorithm an accuracy of 95% in predicting the real volume.

Image segmentation is one of the most difficult problems in multidimensional reconstruction of medical ultrasonic images. A new method for segmentation of medical ultrasonic images based on fractal dimension Lacunarity is presented in this paper. As fractal features have drawbacks of less number and great redundancy, we developed a method to extend the features combining with micromasks. Experimental results show that our method can enhance the description ability of fractal features and segment the images efficiently.

In order to reduce the X-ray dose in the endoscopic examination and theory, we are developing an inside-body navigation system which can give and visualize the direct 3D position of the tip and trace of the endoscope instead of the conventional 2D X-ray fluoroscopy. In this paper, we describe a method of identifying 3D position and direction of ultrasound scanning probe inserted into inside of the body, which gives endoscopic images. We use marker transducers placed on surface of the body, which transmit pulses synchronized to the ultrasound scanner. The method of measuring position (direction and distance between scanning probe inside of body and transducer outside of body) of marker on the scanned image itself obtained with ultrasound endoscope by using adaptive threshold and robust algorithm is described. Further, we give some results of confirmation experiments, and discuss them.

Registration is a problem of motion estimation (ME). The conventional ME methods often can not trade off between the estimation precision and computational complexity. This proposed method based on wavelet transform is a multi-scale block matching one, which takes the full advantage of the self-similarity that occurs after the wavelet transforming. The precision is from coarse to fine, which accelerates the matching procedure and reduces the computation consumption noticeably. This multi-scale registration algorithm has been proved practicable and useful in the real application.

It is difficult to automatically segment and classify tomograph images of actual patient's brain. Therefore, many interactive operations are performed. It is very time consuming and its precision is much depended on the user. In this paper, we combine a brain atlas and 3D fuzzy image segmentation into the image matching. It can not only find out the precise boundary of anatomic structure but also save time of the interactive operation. At first, the anatomic information of atlas is mapped into tomograph images of actual brain with a two step image matching method. Then, based on the mapping result, a 3D fuzzy structure mask is calculated. With the fuzzy information of anatomic structure, a new method of fuzzy clustering based on genetic algorithm is used to segment and classify the real brain image. There is only a minimum requirement of interaction in the whole process, including removing the skull and selecting some intrinsic point pairs.

With the rapid development of aerospace technique, the performance of advanced fighter aircraft becomes improved. These aircrafts can perform beyond the acceleration tolerance limits of human. G-induced loss of consciousness (G-LOC) has become Air-Force's a serious human factor problem. It remains a killer of both men and machines. Various G-LOC research efforts all over the world about the principle and monitoring system have been initiated. Nowadays, in order to detect G-LOC, under a special centrifuge training facility environment, we studied the methods of testing subject's eye blink, head-slump and grip strength measuring force stick and control ability. This article only presents the method for real-time detection of head-attitude, recording head video image any analysis of moving track by video image. We have developed image processing, which is more advanced than infrared technique and laser technique for more information. This method is simple and avoids the affect of infrared ray to partial physiological sensors.

Accurate edge detection of retinal vessels is a prerequisite for quantitative analysis of subtle morphological changes of retinal vessels under different pathological conditions. A novel method for edge detection of retinal vessels is presented in this paper. Methods: (1) Wavelet-based image preprocessing. (2) The signed edge detection algorithm and mathematical morphological operation are applied to get the approximate regions that contain retinal vessels. (3) By convolving the preprocessed image with a LoG operator only on the detected approximate regions of retinal vessels, followed by edges refining, clear edge maps of the retinal vessels are fast obtained. Results: A detailed performance evaluation together with the existing techniques is given to demonstrate the strong features of our method. Conclusions: True edge locations of retinal vessels can be fast detected with continuous structures of retinal vessels, less non- vessel segments left and insensitivity to noise. The method is also suitable for other application fields such as road edge detection.

The goal of this research is to put forward a high efficient method that can generate clinically useful images with improved visualization of retinal vascular image features. Methods: (1) Multiresolution decomposition of an original retinal image into subband images via 2D wavelet transformation (WT). (2) At S equals 2¹, the LLMMSE estimate algorithm that is applicable for nonstationary image model is used in wavelet domain to reduce noise. At s equals 2², soft thresholding wavelet shrinkage technique is used in wavelet domain to reduce noise. At s >= 2³, semisoft wavelet shrinkage technique is applied in wavelet domain to further reduce noise. (3) Opening of a gray-scale A^d_af subband image by a gray-scale circular structuring element can generate a background image. Enhancement is achieved by using this opened image as a unsharp mask and then applying morphological spatial filtering technique to enhance vessel contrast. (4) Reconstruction of retinal image from modified subband images via inverse 2-DWT. Experimental results show that the proposed approach is able to adaptively enhance subtle vascular features, suppress noise and improve global visualization of retinal vascular images.

A major problem in medical image aided diagnostic system is the segmentation of image acquired scenes in order to understand their content through image analysis. This paper presents a fast algorithm for the image segmentation, especially boundary detection of microscope-medical image. The algorithm consists of three steps: (1) reduce the image noise by using a Gaussian-smoothing operator; (2) detect the edge by using a fast method based on fuzzy sets; (3) follow the boundary contour and remove the foreign bodies in the image. The fast algorithm has been tested on real microscope-medical image. The experimental results show that the method is fast and accurate.

Important progresses have been obtained during the last years in the science of processing high resolution radar images. They derive from several facts. At first we have a better knowledge of the signal statistical properties allowing improved performances for the detection and filtering stages. Then, these low level stages have received a great attention and solutions have been found which are specific to the radar images and no longer mimic the processing which proved efficient in the visible range domain. At last, the specificity of the microwave imaging has been taken into account in the objectives of the complete interpretation task.

This paper deals with 3D object understanding from 2D images, including articulated objects in active vision environment. Usually an articulated object is divided into two portions: main rigid portion and articulated portion. It is more complicated than `rigid' object in that the relative positions, shapes or angles between the main portion and the articulated portion have essentially infinite variations, in addition to the infinite variations of each individual rigid portions due to orientations, rotations and topological transformations. A new method generalized from linear combination is employed to investigate such problems. It uses very few learning samples, and can described, understand, and recognized 3D articulated objects while the objects status is being changed in an active vision environment.

RTP/RTCP (Real-time Transport Protocol/Real-time Control Protocol) are independent of the underlying transport and network layers, they are generally implemented based upon datagram-oriented protocols, for a instance, RTP/RTCP are usually implemented based on UDP in a TCP/IP network environment, while based on IPX (Internetwork Packet eXchange) in a SPX/IPX network environment. But as for the situation that only a octet-stream-oriented protocol is provide, the situation that the lost and disorder of data packets are somewhat scarce or the situation that more centralized control is required, it is necessary or practicable to exploit protocols such as TCP which is stream-oriented to implement RTP/RTCP. One of the representative applications is a video conference system based on TCP. As a stream-oriented protocol is employed to implement RTP/RTCP, there is a most major concern is how to accomplish unicast and multicast. This paper present a scheme that use a mixer defined in RTP/RTCP to fulfill multi-point transmission based on octet-stream-oriented protocol. According to this scheme, RTP/RTCP to fulfill multi-point transmission based on octet-stream-oriented protocol. According to this scheme, RTP/RTCP data are transmitted between each station and the mixer while the mixer provides an analysis about RTCP data, and the logical RTP/RTCP framing is performed by each station and the mixer in conformity to the payload type. Considering the characteristic of real-time video data stream, this paper provides a analysis on a exemplary video conference system which uses TCP-based RTP/RTCP to transmit real-time video and audio data. Moreover, a scheme to interconnect between such a octet-stream-oriented protocol based implementation and a datagram-oriented protocol based one is presented, and a discuss about the practicability to combine this implementation to a special simplified RSVP (Resource reSerVation Protocol) also is given forth.

In this paper, a kind of self-similar coefficient (SSC) of the range blocks in an image is defined in Fractal Image Compression scheme. It also is proved that the range blocks in fractal images possess short distance piecewise self- similarity (SDPS). A novel edge detect method is proposed based on SSC and SDPS, and the result show that this method can be used to extract edge of fractal image effectively.

This paper gives a brief description of the method and techniques developed for the modeling and reconstruction of fast moving and deforming 3D objects. A new approach using close-range digital terrestrial photogrammetry in conjunction with high speed photography and videography is proposed. A sequential image matching method (SIM) has been developed to automatically process pairs of images taken continuously of any fast moving and deforming 3D objects. Using the SIM technique a temporal-spatial model (TSM) of any fast moving and deforming 3D objects can be developed. The TSM would include a series of reconstructed surface models of the fast moving and deforming 3D object in the form of 3D images. The TSM allows the 3D objects to be visualized and analyzed in sequence. The SIM method, specifically the left-right matching and forward-back matching techniques are presented in the paper. An example is given which deals with the monitoring of a typical blast rock bench in a major open pit mine in Australia. With the SIM approach and the TSM model it is possible to automatically and efficiently reconstruct the 3D images of the blasting process. This reconstruction would otherwise be impossible to achieve using a labor intensive manual processing approach based on 2D images taken from conventional high speed cameras. The case study demonstrates the potential of the SIM approach and the TSM for the automatic identification, tracking and reconstruction of any fast moving and deforming 3D targets.

Automated quantitative analysis for pneumoconiosis is presented. In this paper Japanese standard radiographs of pneumoconiosis are categorized by measuring the area density and the number density of small rounded opacities. And furthermore the classification of the size and shape of the opacities is made from the measuring of the equivalent radiuses of each opacity. The proposed method includes a bi- level unsharp masking filter with a 1D uniform impulse response in order to eliminate the undesired parts such as the images of blood vessels and ribs in the chest x-ray photo. The fuzzy contrast enhancement is also introduced in this method for easy and exact detection of small rounded opacities. Many simulation examples show that the proposed method is more reliable than the former method.

This paper considers the problem of multi-channel blind image restoration and blur identification. By constructing the blind identification problem into an optimization problem, we propose a subspace decomposition based algorithm to blindly identify the blur functions. The proposed algorithm is inherently the same as many of the others in the literature, but at significantly reduced computation complexity. Let M be the number of blurred images available, N₁ X N₂ be the size of the images and L₁ X L₂ be the size of blur functions, our algorithm has a computational complexity of O(M²L²₁L²₂N₁N₂), as compared to O(M⁴L²₁L²₂N₁N₂) for previous works. The proposed algorithm is therefore more suitable for practical applications.

Increasing or decreasing the internal calcium concentration can promote or prevent programmed cell death (PCD). We therefore performed a Ca²⁺ imaging study using Ca²⁺ indicator dye fura-2 and a sensitive cooled-CCD camera with a 12 bit resolution. Monochromatic beams of light with a wavelength of 345,380 nm were isolated from light emitted by a xenon lamp using a monochromator. The concentration of free calcium can be directly calculated from the ratio of two fluorescence values taken at two appropriately selected wavelength. Fluorescent light emitted from the cells was capture using a camera system. The cell morphology study is based on multispectral scanning, with smear images provided as three monochromatic images by illumination with light of 610,535 and 470 nm wavelengths. The nuclear characteristic parameters extracted from individual nuclei by system are nuclear area, nuclear diameter, nuclear density vector. The results of the restoration of images and the performance of a primitive logic for the detection of nuclei with PCD proved the usefulness of the system and the advantages of using multispectral images in the restoration and detection procedures.

Information fusion by combining multiple sensor data and classifiers has received considerable interest in recent years. It is believed to be an effective way to design pattern recognition systems with high recognition accuracy, reliability, and robustness, to deal with the variance-bias dilemma, to handle different sensor or feature types and scales, and to maximally exploit the discriminant power of individual features and classifiers (experts). This paper provides a brief survey and a taxonomy of various schemes, which are characterized by their (1) architecture, (2) selection and training of individual classifiers, and (3) combiner's characteristics. Recent work on theoretical analysis on combination schemes is also briefly summarized.

This paper presents the theory and formalism of fuzzy causal probabilistic networks (FCPN) and show their current and potential applications in multisensor data fusion. A FCPN is a directed acyclic graph representing the joint probability distributions of a set of fuzzy random variables describing a problem domain. FCPNs extend causal probabilistic networks, also called Bayesian networks, belief networks, or influence diagrams, by associating each discrete variable with a fuzzifier and a defuzzifier, if required. A fuzzifier converts a crisp variable to a fuzzy discrete variable while a defuzzifier does the inverse. FCPNs provide a high-level generic architecture for fusing data incoming from multiple sensors. The paper also provides an overview on the field of multisensor data fusion. Airborne early warning and control using multiple sensors is studied to showcase the theory of FCPNs and their applications for multisensor data fusion.

The basic technologies of designing dual-band imaging system (3 - 5 micrometers , 8 - 14 micrometers ) can be researched in this paper. We study the ratio of the dual-band radiating energy in the different regions of temperature. The relation between S/N ratio and NETD of a dual-band imaging system is discussed and compared with a single band in detail. The S/N and NETD of a dual-band imaging system is 5 times over that of a single imaging system. By calculating and analyzing the imaging characteristic and the dual-band (3 - 5 micrometers , 8 - 14 micrometers ) detectors with the same focus plane and size, the fusion technology in level of pixel is obtained and realized using main-self fusion structure, which the processing and complexity of a dual-band can be reduced.

Data fusion is heavily concentrated in both inland and overseas. Nowadays, among many developed algorithms, Dempster-Shafer reasoning and fuzzy set technology are widely used and most representative. A new data fusion approach of generalizing Dempster-Shafer reasoning to fuzzy set is presented in this paper. And it is used to deal with by object recognition with rosette scanning dual color data from infrared and ultra-violet sensors. The simulated experiments illustrated that our approach can eliminate false targets and recognize target correctly.

This paper presents a fuzzy-neural network with a fast global optimization algorithm and its application in the data fusion. The fuzzy-neural network is based on the global property of genetic algorithm and the high speed property of expectation maximization algorithm. Simulations results of its application show that this fuzzy-neural network is of robustness in the data fusion.

A simplified system to fuse the infrared and visible light images in real time is introduced in this paper. A Thermal Image System and a CCD camera are used to obtain the infrared and visible light images respectively. The two different sources images can be correlated completely both in the time and in the space after we solve the problems of synchronizing, dimension matching and visual field matching. Afterward, these images can be used in real time fusion or used the further fusing algorithm.

Imaging spectrometer is one important aspect of contemporary remote sensing developments. It is characterized by huge amount of supermultipal spectral image data provision. For less remote sensing image data provision, we used to put all the data directly into the process of image object recognition without making any data fusion. Even occasional data fusion is experiential. This situation is not suitable for imaging spectrometer. The supermultipal bands of imaging spectrometer provide us with much more freedom on available choices of suitable images for procession and information extraction; at the same time, fusion between different related image data aimed at themes extracting can reduce the amount of data on one hand, and make most use of imaging spectrometer image data on the other. And to some extent, image data fusion is of utmost importance and practical significance for theme information extracting from imaging spectrometer images. This paper discusses fusion methods for conventional remote sensing data. Based on analysis of imaging spectrometer, it proposes method for imaging spectrometer image data.

Diagnostic features in mammograms vary widely in size and shape. They are of low contrast, and pathological features in mammograms are often vague and very hard to discriminate. This paper presents a novel approach for segmentation of suspicious mass regions in digitized mammograms, using logic filters. First we enhance the edges of suspicious region in mammogram using one kind of logic filter. The image result gives us a special behavior on the gray level histogram, which facilitates the segmentation of the suspicious mass. The performance of our method is shown using 25 digital mammogram images. A comparison with other methods is also given briefly.

This article introduces a model of MPP system by the aspects of overall machine architecture, processor interconnection net, algorithm model. This system adopts some basic design principles of MPP system, while it has its own characteristics in concrete structure. For example, in the overall machine architecture MPP system is used as co- processor and a high performance personal computer is used as host machine; the system adopts distributing memory principle, uses register as local memory, uses column buffer as the adapter between shared memory and processor array. This system is mainly used for developing data parallelism. Under the CU (Control Unit) centralized controlling, all processors execute same instruction. This kind of structure is very suited for different kinds of image processing, such as convolution, transformation and median filtering. Another characteristic of this system is scalability. It can dynamically expand with question's size in logic and physical sense. In addition, this article uses two laws of parallel processing. Amdahl law and Gustafson law, to explain why the MPP system is suited for primitive image understanding.

This paper will describe the embedded SIMD massively parallel processor that we have developed for real-time image processing applications, such as real-time small target detection and tracking and video processing. The processor array is based on SIMD chip BAP-128 designed by our own, and uses high performance DSP TMS320C31, which can effectively perform serial and floating point calculations, as the host of the SIMD processor array. As a result, the system is able to perform a variety of image processing tasks in real-time. Furthermore, the processor will be connected with a MIMD parallel processor to construct a heterogeneously parallel processor for more complex real- time ATR (Automatic Target Recognition) and computer vision applications.

Discrete cosine transform (DCT) is widely used in signal processing. This paper presents a novel approach to perform DCT. DCT is expressed in terms of discrete moments and a systolic array for computing DCT with only a few multiplications and without any cosine evaluations has been proposed. The execution time of the systolic array is only O(Nlog₂N/log₂log₂N) in computing 1D DCT. The approach is also applicable to multiple dimensional DCT and DCT inverses.

The architectural design of an image processing system and its instrumentation is elucidated here in this presentation, based upon the multiprocessing TMS320C82 DSP processors, working under a specially-defined MAEVRISC-DSP (Multidimensional Application-Embedded Vector RISC-DSP) architecture, that blends high-precision and high- performance (120 MFLOPS digital and 100 MSPS analog) data acquisition with efficient signal processing architecture design and prevalent Tagged MRI pulse sequence algorithms, specifically optimized for the medical imaging applications of Magnetic Resonance Spectroscopy and Imaging (MRI) tomography, while at the same time giving a low-cost alternative to extremely-expensive MRI systems. Alongside the system definition, the design of a new magneto-optical instrumentation is also presented, named SQUID and LPDA- based Field Equalization and Susceptibility Detection sensing, which works upon the recently-defined Tunneling Photon Resonance effect, mainly devised with non-ionizing human brain tumor diagnosis and localization in perspective, whose brief account is highlighted here.

Moment invariants, by virtue of invariance properties under translation, scaling and rotation, have played an important role in pattern recognition and computer vision. In the present paper, extending the systolic array for computing moments, a global systolic structure for fast computation of moment invariants has been proposed. This structure fully exploits parallelism of the network architecture and reduces computational time complexity from O(n²) in the case of direct computation to O(n) (n denotes image size). The systolic structure nearly consists of adders with area complexity O(n) and are highly regular and structurally very simple, resulting in simple hardware implementation. The method is suitable for both binary images and gray level images and is also suitable for image sequence moments.

Though it promised a high compression ratio and high decompression quality, the low fractal image coding speed blocked its way to practical application. The main problem of the low coding speed is due to the long searching time of matching between domain blocks and range blocks. Genetic algorithms (GAs) are well known as a search and optimization technique based loosely on the principles of natural selection. In order to decrease the coding time, here a method is proposed that employs genetic algorithms to solve the matching problem of fractal image coding. Experimental results indicate that the introduction of GAs to fractal image coding is a good way to increase the speed of fractal image coding without decreasing the image quality obviously.

3D imaging is the science of the methods of defining, visualizing, manipulating, and analyzing 3D object information captured in multidimensional images acquired from multiple modalities. The purpose of this paper is to give an overview of this science and to identify the major challenges currently faced in its medical applications. It describes the 3D imaging operations under four groups: preprocessing, visualization, manipulation, and analysis. Preprocessing operations go toward defining objects in given images. Visualization operations facilitate seeing and comprehending objects in their full dimensionality. Manipulation allows altering object structures and relationship among objects. Analysis deals with methods of quantifying object information. The paper describes the concepts underlying these operations. It draws two main conclusions: (1) There are two classes of approaches to 3D imaging--those based directly on image intensities and those based on object shape. (2) There are two main groups of challenges, related to object delineation and validation.