The convex feasibility problem of finding a point in the intersection of finitely many nonempty closed convex sets in the Euclidean space has many applications in various fields of science and technology, particularly in problems of image reconstruction from projections, in solving the fully discretized inverse problem in radiation therapy treatment planning, and in other image processing problems. Solving systems of linear equalities and/or inequalities is one of them. Many of the existing algorithms use projections onto the sets and may: (i) employ orthogonal-, entropy-, or other Bregman-projections, (ii) be structurally sequential, parallel, block-iterative, or of the string-averaging type, (iii) asymptotically converge when the underlying system is, or is not, consistent, (iv) solve the convex feasibility problem or find the projection of a given point onto the intersection of the convex sets, (v) have good initial behavior patterns when some of their parameters are appropriately chosen.

In this paper, a PUMA562 robot, a micro CCD camera and an image acquisition card, along with IBM/PC 586 host computer, formed eye-in-hand visual servoing experimental system of an intelligent space robot. A novel robot visual servo control method is adopted, based on FMF neural network. Computer simulation and many experiments of object tracking with the capture of floating object were carried out in this eye-in- hand visual servoing system. It is an important reference to further research in the future.

In cone-beam 3D CT, 2D X-ray projections determine the reconstruction quality directly. However, the generation of an X-ray projection is a complex process that includes the emission, the conversion and the transfer of X-ray latent information, so the pixel value of a projection has no exact meaning generally. In this paper, we suggest a way by experiments to model the whole imaging system based on 'equivalent single energy' which is used to analyze polychromatic X-ray. We define each tissue's SCF(single calibration factor) of human and get GCF (gross calibration factor) by summing up every weighted SCF, then we can use GCF to calibrate and linearize X-ray projections. After such processing, X-ray projection will have explicit meaning, i.e. its pixel value can denote the equivalent material thickness linearly, which will be advantageous to the algorithm research of 3D reconstruction and other quantitative analyses of human's physiological parameters.

A noel approach to reconstruct shape from shading information is presented, which considers both the global and local shading. Firstly, a 2D gray-level image is divided into several simple patches with snake model. Secondly, the shape of the patches is reconstructed respectively. Thirdly, the different patches are jointed into a whole one. And at last, the mosaic surface is smoothed. According to this method, not only the distortion caused by image discontinuity but also the effects of local noises is eliminated. Results of experiments are given to illustrate the usefulness of the approach.

In this paper, we apply the geometric curve evolution approach to 3D reconstruction for a number of images in the stereo vision. The curve evolution approach is based on the Euclidean curve shortening evolution theory, and the level set method is introduced into it for numerical computation. The Euler-Lagrange equations that are deduced from the variational principle provide a set of curve evolving equations. These PDE's describe the process of the geometric curve evolution on the relevant epipolar plane. In 3D space, the each epipolar plane is monogamously projected as a unique pair of the intra-scanlines on stereo pairs. These PDE's are used to deform an initial set of curves that then move towards the object outlines to be detected. The level set implementation of these PDE's provides an efficient and robust computational way for the kind of the geometric-driven evolving equations. The velocity term in the level set equations can be obtained from the above geometric curve evolving equations. It is intrinsic and only depends upon the stereo problem. We present the close form of the velocity. Finally, the results of implementation of our theory are presented on synthetic images.

Directional effects of illumination make difficult classification of rough 3D surfaces as their appearance may change dramatically, especially when a surface has variable albedo. One way of circumventing the problem is to separate the local shape and albedo information prior to classification. We do it by means of Colour Photometric Stereo, which produces 5 descriptors for each surface patch: two gradient components, and three colour components. This information is illumination-invariant, and can be used as an input for a suitable classification scheme. We proceed to classify a collection of surfaces by matching their colour histograms and multidimensional co-occurrence matrices of shape descriptors.

A new algorithm for the reconstruction of tomographic images from a few views is presented. A variable metric method is used to solve the unconstrained optimization problem that resulted from the analysis by the use of the maximum entropy formalism. The numerical simulations study the reconstruction effect on the different asymmetric functions and the stability against noise. The results show that the reconstruction accuracy is adequate.

In this paper, a 3D reconstruction approach based on wavelet analysis is presented. It can be used in neuro-vision system. The approach can be divided into two parts. First, the stereo matching problem is solved with wavelet analysis. Dyadic discrete wavelet analysis is adopted in this process and stereo matching process is realized with global optimization. A coherent hierarchical matching strategy is constructed, so that the stereo matching process can be accomplished with coarse to fine techniques. Second, a 3D object reconstruction neural network is constructed by using BP neural network. By feeding the image corresponding points between the left image and right image in a stereo image pair, the 3D coordinates of points on object surface can be obtained using this neural network and the configuration and shape of the object can be reconstructed. With multiple 3D reconstruction neural networks the 3D reconstruction processes can be performed in parallel. The examples for both synthetic and real images are shown in the experiment, and good results are obtained.

In this paper we propose a new approach to address the 3D multi frame reconstruction problem, by which a high resolution image is reconstructed form several low resolution images. The originality of this approach is that we regard the 3D perspective transformation as a part of the degradation faced by the imaging process from the original 3D object surface to the 2D imaged frames. Therefore, we reconstruct the pixel values of the original image directly from the low resolution image sequences. The quality of this approach is well demonstrated by simulated experiments.

It is a hotspot problem in robotics research that determine the related pose between two space objects and realize their cooperate. 3D visual method is the main method for measures the space object pose. Because the rigidity demand during pose computes, the 3D vision at this condition ins model based. That is to say the model of the marks must be used as a strict restriction during the computing. This paper will give the method of computing 3D pose at first and discuss the model restriction problem emphatically, include model- based monocular vision method and model-based binocular stereo. The compare analysis between the two methods and the experiment result will be given at last.

This paper proposes a calibration system consisting of three components: a quasi-linear intrinsic calibrator, a linear extrinsic calibrator and a nonlinear L-M optimizer. The focal length is evaluated form vanishing points. Then the rotation matrix and the translation vector are estimated linearly. At last a Levenberg-Marquardt optimization is performed to refine the extrinsic parameters by minimizing the reprojection error. The parameterization of the rotation matrix is discussed in detail, and two parameterization methods, Euler-Angle and Axis-Angle are compared. Experimental results prove that the system can calibrate the cameras precisely.

In this paper, we implemented a three dimensional image display system using stereogram and holographic optical memory techniques which can store many images and reconstruct them automatically. In this system, to store and reconstruct stereo images, incident angle of reference beam must be controlled in real time, so we used BPH (binary phase hologram) and LCD (liquid crystal display) for controlling reference beam. And input images are represented on the LCD without polarizer/analyzer for maintaining uniform beam intensities regardless of the brightness of input images. The input images and BPHs are edited using application software with having the same recording scheduled time interval in storing. The reconstructed stereo images are acquired by capturing the output images with CCD camera at the behind of the analyzer which transforms phase information into brightness information of images. The reference beams are acquired by Fourier transform of BPH which designed with SA (simulated annealing) algorithm, and represented on the LCD with the 0.05 seconds time interval using application software for reconstructing the stereo images. In output plane, we used a LCD shutter that is synchronized to a monitor that displays alternate left and right eye images for depth perception. We demonstrated optical experiment which store and reconstruct four stereo images in BaTiO3 repeatedly using holographic optical memory techniques.

3D image processing is a subset of computer vision area. Object pose estimation is an important application in 3D object recognition. A 3D object pose estimation method is developed for an automated manufacturing assembly application. The target parts are extracted from the original range images using the traditional edge detection and segmentation methods. The center position is then computed through the circle Hough transform algorithm. For the 3D orientation estimation, a 3D geometrical feature descriptor, angle distance map (ADM), is proposed to describe the 3D local surface feature. A triangular mesh model of 3D object is used for reducing the computational complexity. The principal components analysis method is applied on the ADM descriptions for efficient comparison. The orientation information is computed according to the extracted 3D feature points. The proposed method is tested in an application for flexible robot assembly. The experimental results show that accurate 3D pose estimation can be obtained.

A new method for reconstruction of surface from cross- sectional contours is presented base don iterated function system . The properties of the algorithm are, nonlinear and stochastic. The experiment result shows that the prosed scheme can accurately preserves the shape of the contours with a very fast calculation speed.

In the paper, we present a new surface rendering method using a modified contour matching algorithm. The algorithm is based on the invariance of curvature after rotating or displacement. Using the description of contour by curvature, the problem of contour matching in 2D is converted into a 1D problem. Therefore, it decreases the computational cost. In the computing, we compared two kinds of numerical algorithms, i.e. predictor-corrector scheme and successive over relaxation (SOR). Finally, we introduce a 3D surface rendering procedure employing the contour matching.

The principle of projection structure light for 3D vision of small objects is described in this paper. Three vision systems, projection fringe, projection speckle and a special ring light illumination system, are introduced as the application of the principle. The ring light illumination system is specially designed for the 3D vision of small objects with specular surface like BGA balls. The testing result listed in the paper reveal that these methods can satisfy the requirements of 3D vision of small objects.

This paper presents a new effective object orientation method using hand-eye vision system. First, we set a constant transformation matrix to describe the fixed correspondence of the end-effector reference frame and the robot-base reference frame. The constant matrix can then simplify the 3D orientation problem into 2D orientation problem in the case of keeping constant height while calibration. It greatly reduces the computation and increases the accuracy. When the position of end-effector has been changed, it can still implement orientation without the need of re-calibration. The experiment results show that this new method is highly accurate and reliable.

Video segmentation, key-frame selection and description are essential steps for CBIR. New techniques about video sequences to suit the needs of CBIR are studied in this article. Data structure of video, a key of CBIR, is defined by scene, shot and key-frame of shot. Secondly, we propose a completely implement method of scene segmentation and shot segmentation using phase correlation function of differential images. Moreover, we define key-frame of a shot base don the mosaic picture an describe it based on the Fourier Descriptor after we discuss various select schemes of shot key-frame depend on the characters of video sequence. There are diverse movements between two frames, such as shift, rotation and scale. In this paper, we introduce the phase correlatively between images in log- polar coordinates to detect rotation and scale and the phase correlativity in Cartesian coordinates is exploited to find shift. We further demonstrate that the proposed method as a powerful tool for segmentation of video sequences.

Generating semantically meaningful content of news video has been increasingly spotlighted. Captions in video include useful information for automatic annotation and indexing. Unfortunately, there are difficulties in retrieving captions for most real applications because of the following factors: low resolution characters and extremely complex backgrounds. This paper proposes a novel local mutation-based method to solve these problems. We detect and locate captions according to shape-regularity and connectivity of the local mutation region. The overall experimental results show that our approach is general, effective, real time and robust enough for caption detection and retrieval, and farther for use in news indexing.

With the wide application of digital video, video database system has been an important direction in multimedia domain. In the field of video database research, segmentation of a video scene has been an important problem hard to solve. This paper begins with analyzing the structural features of news video and then introduces a new method for scene segmentation based extracting the semantic features. In this method extraction and analysis for shot change, anchorperson shot, topic caption and silence interval in audio and video stream are carried out to realize news video scene segmentation. It can be seen from the result of experimentation that the correctness to segmentation rate is up to 86.9 percent with the application of the method mentioned above. Hence, the method is a satisfied solution for the news video scene segmentation.

Examining wheel images is needed for train's safe running. Based on computation of image similarity, a method is introduced in this paper to pick a few available images for examination from a series of train's images, in which containing whole and clear wheels. First, histogram is calculated and histogram intersection is adopted to evaluate the histogram similarity between standard image and candidates, then a multi-resolution decomposition using Haar wavelet is applied, several edge statistical features based on wavelet coefficients are obtained and similarity comparison on the features is processed. And then these two kinds of similarity are combined to synthetically express the similarity between two images. Finally, those images satisfying threshold are picked for examination. Experiments show useful wheel images can be selected efficiently in this method.

Presently, image sequences segmentation algorithm can be mainly separated into two parts. One is based on brightness, chroma and margin pixel information, the other is based on frame disparity information, just like Frame Disparity Threshold, Change Detection Mask, High Order Statistic and so on. The first method is seldom used recently, while the latter one is deficient in noise-sensitive. So, we take a special point of view in this paper, and present da new segmentation algorithm based on wavelet with timeline method. Here timeline is offered to control time sequences. Firstly, we transform the image sequences by wavelet on the timeline. After the transformation, we should hold the high- frequency coefficient on the part of motion, and then we obtain motion object's mask by morphological process. By such a series of operations, we can get the final motion object. Finally we device some experiments to measure the methods processing efficiency and real-time properties. The results show that the method is simple and practical.

The estimation of rigid motion parameters has been investigated using Fourier transform and iterative techniques for over two decades. But the numerical efficiency remains a challenge. Especially for 3D case, there is no analytical algorithm. In this work, we propose to use the moment covariance matrix to estimate the orthogonal transform (i.e. rigid motion in real world). This algorithm does not depend on the dimension of the problem and is an analytical method. The numerical implementation is easy and efficient. The test results show that the proposed algorithm is accurate and well tolerent to noise variation.

Caption graphically superimposed in video frames can provide important indexing information. The automatic detection and recognition of video captions can be of great help in querying topics of interest in digital news library. To detect the caption from video sequence, we present algorithms based on fuzzy clustering neural networks. Since neural networks have the capabilities of learning and self-organizing and parallel computing mechanism, with the great increasing of digital images and video databases, neural networks based techniques become more efficient and popular tools for multimedia processing. Experimental results show that our caption detection scheme is effective and robust.

This paper presents a new method for image coding information processing based on CDMA, where can use the PN sequences applying to multispectral. There are two families of orthogonal coding sequences have been used. Also hiding a digital image information technology is introduced.

Illumination variance will reduce the correctness of motion estimation result. To improve ME algorithms robustness, some algorithms have been proposed. All those algorithms are based on the intensity illumination models and many important color information are ignored. Therefore, in this paper, starting from the color reflection model, we propose a novel illumination invariant feature named NACD. Based on NACD, the improved ME algorithm, FC-ME, is designed to deal with the illumination variant image sequences. The experiments indicate that with the acceptable additional operations, the new algorithm's robustness to illumination variance is improved obviously.

Virtual Endoscopy System is a new aided diagnosis method based on computer processing of 3D image slices to provide simulated visualizations of specific organs similar to those produced by standard endoscopy. Compare with real endoscopy, VES has much advantages and will have more applications in the future. We constructed a Virtual Bronchus Endoscopy System based on the techniques of image analysis, compute graphics, and so on. Based on the characteristic of bronchus, we adopted an improved 3D region-growing algorithm, which we call 3D scanline algorithm to extract the bronchus from the DICOM-formatted medical images, then the 3D polyhedral surface model of bronchus is obtained by triangulation with Marching Cubes Algorithm. Then the user is allowed to navigate freely inside the bronchus along the axis. We adopted surface rendering method in the rendering process. In application this system can meet the requirement of real-time navigation and has pretty good display.

We establish an IR model for ground target such as tank considering the effect of inner heat source, friction and various environmental factors and draw IR images of target at various states using the techniques of Computer Graphics. We then propose a thermal model for different grounds using 1D finite difference method to calculate the surface temperature changing with time. The result are quite satisfied compared with the measured values. Also the re- distributing of temperature field after the touch between targets and ground is calculated and the realistic IR synthesis images of targets and the ground background are drawn. Finally by using the methods of 3D finite difference, we successfully calculate and render RI shadow which describes the effect of higher temperature distribution of the center area originally occupied by targets than that of the around area after the targets are driven away. We also draw different IR images of the ground at the different intervals after the targets have been driven away. Based on these IR images, we can determine the existence of the IR targets some certain interval ago.

This paper concentrates on creating individual faces for MPEG4 compliant animation with triangular B-spline. We reconstruct 3D face by adapting a generic face model. A semi-automatic method, which creates human face from Cyberware range and texture dat, and deformation primitive, an auxiliary tool for face editing, are presented. In both tools triangular method more robust to range data errors and facilitates the optimization process. For the Cyberware data of an individual face, our method first modifies the generic model to initialize in a semiautomatic way. Then non-linear optimization is performed to decide shape information for this specific face. Texture mapping is trivial after face shape is decided. Our method needs a minimum user interaction. It is also robust for different initialization. Our deformation primitive, which is designed semantically, facilitates face-editing procedure greatly. The reconstructed 3D face can be animated immediately with MPEG4 Facial Animation Parameter streams.

As to the research of 3D reconstruction in modern medical imaging, if an object to be reconstructed can be simulated by computer, its X-ray projections will be gotten exactly and freely, which will be more significant in testing the behavior of reconstruction algorithms. In this paper we present a new, efficient method to simulate coronary arteries under the background of cone-beam reconstruction. In order to achieve this task, we create a bifurcate rule-base of coronary arteries and model their spatial configuration by controlling the parameters in spherical coordinates, then we use random iterated algorithm to simulate them by the reference of fractal theory. As to the research of 3D reconstruction in modern medical imaging, if an object to be reconstructed can be simulated by computer, its X-ray projections will be gotten exactly and freely, which will be more significant in testing the behavior of reconstruction algorithms. In this paper we present a new, efficient method to simulate coronary arteries under the background of cone-beam reconstruction. In order to achieve this task, we create a bifurcate rule-base of coronary arteries and model their spatial configuration by controlling the parameters in spherical coordinates, then we use random iterated algorithm to simulate them by the reference of fractal theory.

The virtual studio concept replaces real background sets with computer-generated synthetic scenes. In this paper, we discuss the optical landmark-based camera tracking method in virtual studio applications. Studio cameras are zoom lens imaging device, both the internal parameters and external parameters of the camera are allowed to vary from image to image. Optical tracking --uses pattern recognition--eliminates the need for painstaking calibration of the lens system and can be used with any camera mount, including hand-held cameras. Here we address the problem of accurately tracking the 3D motion and focus of a monocular camera in a known 3D environment and dynamically estimating the 3D location and focus of the camera. We utilize fully automated landmark-based camera calibration to initialize the motion estimation and employ extended Kalman filter techniques to track landmarks and to estimate the camera location and focus. The implementation of our approach has been proven to be efficient and robust and our system successfully tracks in real-time at approximately 25 Hz. This paper describes several years of work at the NUDT Multimedia laboratory to develop optical pattern recognition and tracking systems for use in virtual studio.

This paper presents a general quantitative approach to performance evaluation of ATR systems, especially for FLIR imagery. The general quantitative evaluation approach we proposed consists of three main steps. First, multiple sets of testing image sequences with significantly varying difficulty for specific environment, which can be precisely measured, are transmitted into an ATR system through a general protocol interface. Secondly, the computerized results of each set of testing image sequence are gotten through the interface after the ATR system works on each received image. Finally, data analysis will be prosecuted on the former experimental data sets. For convenience of arrangement, the different sets of testing image sequences with measured parameters and their homologous results are managed as the form of testing groups. When evaluating an ATR system, we design numerous groups of testing data which grade from easy to hard. We have developed the General Evaluation Software for Performance of ATR Systems (GESPATRS), which running under Windows in PC. As an example, an ATR system for FLIR imagery is evaluated by the GESPATRS. More details about it are given in the paper to demonstrate the feasibility of the proposed approach.

We present a 3D reconstruction method, which combines geometry-based modeling, image-based modeling and rendering techniques. The first component is an interactive geometry modeling method which recovery of the basic geometry of the photographed scene. The second component is model-based stereo algorithm. We discus the image processing problems and algorithms of walking through in virtual space, then designs and implement a high performance multi-thread wandering algorithm. The applications range from architectural planning and archaeological reconstruction to virtual environments and cinematic special effects.

Spectral image sensors provide images with a large number of contiguous spectral channels per pixel. Visualization of these huge data sets is not a straightforward issue. There are three principal ways in which spectral data can be presented; as spectra, as image and in feature space. This paper describes several visualization methods and their suitability in the different steps in the research cycle. Combinations of the three presentation methods and dynamic interaction between them, adds significant to the usability. Examples of some software implementations are given. Also the application of volume visualization methods to display spectral images is shown to be valuable.

Signal processing algorithms (SPA) play a key role in an imaging IR tracker which is widely used in infrared search and track (IRST) system. When being used to target detection, recognition and tracking, SPA has a significant influence on the performance of the IRST system. Due to the variety and complexity of field scenes and countermeasures, SPA should be robust enough for military use. The question arises as to how to measure and assess the performance of SPA efficiently and properly. On the other hand, the study on SPA performance evaluation will not only give an assessment to a specific algorithm, but also help the development of signal processing technology. In this paper we propose a way to set up an evaluating system in the SPA research laboratory and study the criteria for the evaluation.

The ability of accurately and efficiently planning safe routes that achieve mission goals in a timely fashion is a requirement for many diverse applications, including military operations. Standard route planning algorithms usually generate a minimum cost route based on a predetermined cost function. Unfortunately, such a route may not represent a desirable solution for different military scenarios. In this paper, we present a 3D route planer for unmanned air vehicles based on Lagrange Programming Neural Networks. Our approach can efficiently handle various constraints on the resultant route, such as minimum route leg length, maximum route length, maximum turning angle, maximum climbing/diving angle, minimum flight height.

This paper presents a fast algorithm for real time image rotation. This fast algorithm has many benefits, of which its can reduce computation complexity and easily be implemented in hardware. The system using this algorithm has many advantages over optical image rotation system. It has low volume, low cost and high reliability.

Silver halide emulsion is an important component of image materials. They are widely used in many kinds of fields and have a very prosperous prospect. This kind of method provides a new way to research the photosensitive process of image materials. Compared with the traditional methods, it is more economical and has a shorter experimental period. In this paper, Nucleation and Growth model and Monte Carlo method were adopted to simulate the formation of image and results in different kinds of outside conditions and material properties will be given.

Data mining usually means an efficient discovery of knowledge from databases, and the immune algorithm is a biological theory-based and globally searching algorithm. The aim of applying immune concepts and the concerned theories is mainly to utilize the local information to intervene int eh globally parallel process, restrain or avoid repetitive and useless work during the courses, so as to overcome the blindness in action of the crossover and mutation. Three important data mining issues addressed by the algorithm are the interest of the discovered knowledge, the computational efficiency of the algorithm, and the trade-off between expressiveness and efficiency.

This paper evaluates five potential feature selection methods in the application of remote sensing. The five methods include the sequential forward floating selection, the joint pair approach, band selection based on variance, the principal component transform, and the visual-based selection. Optical-sensor image and synthetic aperture radar image are used for experiments. Several recommendations are made based on this study. For optical-sensor images, the five feature selection methods: sequential forward floating selection, joint pair, band selection, principal component transform, and visual-based selection could have about the same classification accuracy using two to five selected features. The case study has shown that the sequential forward floating selection is the best feature selection method for both optical and synthetic aperture radar feature image selection, followed by the joint pair (for two-feature selection), visual-based selection, band selection, and principal component transform. For band L and band X synthetic aperture radar feature images, entropy, homogeneity, inverse difference moment, and maximum probability, East to West and West to East semivariogram, the local mean value, maximum, and minimum are the best features of the co-occurrence matrix model, semivariogram model, and local statistic model. For Landsat TM images band 7, 4, 5, 3, 1, and 2 are significant feature images. Applying the sequential forward floating selection to select two to five features from the potential features can obtain classification accuracy greater than 90 percent.

This paper first presents a new scientific accuracy measure (denoted by G) for assessing/evaluating the performance of computer medical diagnostic (CMD) systems by incorporating the true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN) of human and computer's diagnoses with respect to each other. Based on G, a formula for computing a multi-parameter sensitivity vector S(G), with the assumption that the system parameter percentage variations are small, is then proposed. For a given set of parameter percentage errors, from the expression of S(G), we can compute the error bound of G and assess the reliability of the system with human and/or computer errors being taken into consideration. It has been demonstrated that the new measure G is capable of providing consistent performance evaluation of a CMD system in general. Based on the value of G, a CMD system can be classified as having 'good', 'fair', or 'poor' performance. Even though the proposed basic accuracy measure and its sensitivity study are derived based on the diagnosis using two diagnostic categories (positive and negative) compared by two observers (a human expert and a computer system), however, its methodology can be extended to CMD systems with multiple diagnostic categories and observers. The formulas for measuring the performance of such systems are discussed and present.

In the corse of completing the hierarchical reconstruction, recovering projective depth is the key step. The existence methods are very efficient for simulation data, but they are not perfect for real image. In this paper, the estimation of projective depths based on genetic algorithm is proposed, by which the projective depths are iteratively estimated so that the measurement matrix is made to be as close as possible to rank 4. The validity of the proposed algorithm is confirmed by experiments.

Combining the active digitizing technique with the passive stereo vision, a novel method is proposed to acquire the 3D data from two 2D images. Based on the principle of stereo vision, and assisting the active dense structure light projecting, the system overcomes the problem of data points matching between two stereo images, which is the most important difficulty occurring in stereo vision. An algorithm based on wavelet transform is proposed here to auto-get the threshold for image segment and extract the grid points. The system described here is mainly applied to digitize the 3D objects in time. Comparing with the general digitizers, it performs the translation from 2D images to 3D data completely, and gets over some shortcomings, such as slow image acquiring and data processing speed, depending on mechanical moving, painting on the object before digitizing, and so on. The system is the same with the non-contact and fast measurement and modeling for the 3D object with freedom surface, and can be employed widely in the fields of Reverse Engineering and CAD/CAM. Experiment proves the efficiency of the new use of shape from stereo vision (SFSV) in engineering.

This paper presents a hybrid Bayesian approach based on MRF/GRF and active contour models for disparity estimation and segmentation using stereo images. Smooth and accurate disparity field is obtained by using hierarchical MRF and GRF models. In the procedure of disparity estimation, hierarchical overlapped block matching and a fast search method are incorporated to improve the precision and computation. Then pixel-wise refinement is performed on the initial disparity field with edge information to get a smooth and consistent disparity field with sharp boundary. Finally, the active contour model is used to extract the disparity contours by jointly exploiting the information of edge and disparity. The resulting disparity field and corresponding contours are very useful in object-based stereo image coding and object segmentation. Experimental results illustrate the performance of the proposed algorithm.

Parametric curve approximation technique is frequently used in many fields of computer science. The most difficult problem in parametric curve approximation is parameterization. In this paper an Iterative Closest Point reparametrization strategy is presented to improve the precision of approximation. Moreover, this paper proves the local convergence of the algorithm, which makes it more theoretically reasonable and robust. Some results comparing with traditional approaches are given to illustrate the validity of our approach.

Gaskets can be derived from the iterated function systems, which are called fractals due to the random choice of the iterated function systems. The chaotic behavior of iterative procedures has been the subject of many approaches. Utilizing the chaotic characteristics of local unstable and sensitive to the initial condition, this paper gives a simple method using the modified logistic-map to generate beautiful gaskets.

This paper presents a new ways-free image registration method based on adherent mark recognition method. Accurate 3D coordinates are first obtained in each unit-measurement and then all the unit-data information are amalgamated into a communal coordinate system. The stereo vision based 3D profilometer computes at least three marks' 3D coordinates first and then these coordinates are applied to calibrate the translation and orientation matrix between the two different inspection positions. This method allows any freedom of translation or orientation between every two unit-measurements without special constraints. With all of the six possible freedom transformations, the system's flexibility and adaptability are greatly enhanced, so the objects under inspection can be measured according to their inherent shapes at each optimal angle of view. As there needs no redundant ways in the measurement, the portable on- spot inspection becomes feasible. The paper analyzes two different amalgamation algorithms in detail and finally computer simulation results are given.

In this paper, we propose a flexible new technique to calibrate a camera easily. It is well suited for use without specialized knowledge of 3D geometry or computer vision. The technique only require the camera to observe a planar pattern shown at a few different orientations. Either the camera or the planar can be freely moved. And the motion need not be known. The whole procedure can be done automatically. Both computer simulation and real data have been used to test the proposed technique, and prove that the method has some effect.

3D surface reconstruction has numerous applications. Many effective algorithms have been developed in the past 20 years. In this paper, we introduce a new surface reconstruction algorithm based on degree of neighbors, which is defined by MST (Minimum Spanning Tree) and compensated scheme. It makes the parameter more stable and easier to be chosen experimentally. Moreover, it steers the reconstruction to be more applicable for sparsely, unevenly sampling points by greatly reducing the disturbance of non-topological neighbors. The massive results of examples illustrate that our approach is effective and competitive.

Based on the analysis on the bottleneck for improving CCD measurement resolving power in the high-accuracy image detection system, sub-pixel subdivision technology is presented. Through comparison the advantage and disadvantage of current subdivision algorithms, we propose a new 1D sub- pixel subdivision algorithm called Adaptive Weighing Fitting algorithm in the paper, which is suitable for such image as low contrast, low SNR and complex background. And we implement the real-time algorithm in the high-speed system platform based on TMS320C50. Then a fast 2D sub-pixel subdivision algorithm based on correlation feature and direction feature of image border is also discussed for improving the location accuracy of complex border. By many simulation experiments and practical detection in spot, the satisfied subdivision result are obtained.

In this paper, we describe some new methods to perform color CCD camera calibration in virtual studio. Our work is based on the pinhole camera model with lens radial distortion. Generally, the calibration problem consists of two steps: control points extraction from images and calibration algorithm implementation. For control point extraction, the usual way is relatively simple and can be easily done. But because of the poor quality of grabbed images and the lens distortion, the usual method has two main disadvantages in our virtual studio application. Hence we present an effective and robust method to perform control points extraction. Generally, the calibration experiments are implemented by Tsai's two-stage algorithm. But in our experiments, we find that the calibration problem is ill-conditioned, for the result is very sensible to tiny changes of calibration points' coordinates. To solve this problem, we present a new strategy to improve the accuracy of calibration results of focal length and depth. Then we obtain a focal length look-up-table (LUT) relative to the lens zoom rings. In virtual studio application, when the real camera pans or tilts, the background picture lags behind the foreground picture. Then a motion compensation model is presented to compensate the difference between the optical center and the real rotation center to overcome the problem.

The solution of camera's orientation parameters is one of the most important tasks in photogrammetry. The approach presented in this paper is based on the coplanar relationship, three-axes orthogonal constraints and parallel line information. The interior orientation parameters and rotation matrix are solved with parallel information and least-square adjustment. The test of the prosed method is based on the real data.

In this paper, firstly, several video shot detection technologies have been discussed. An edited video consists of two kinds of shot boundaries have been known as straight cuts and optical cuts. Experimental result using a variety of videos are presented to demonstrate that moving window detection algorithm and 10-step difference histogram comparison algorithm are effective for detection of both kinds of shot cuts. After shot isolation, methods for shot characterization were investigated. We present a detailed discussion of key-frame extraction and review the visual features, particularly the color feature based on HSV model, of key-frames. Video retrieval methods based on key-frames have been presented at the end of this section. This paper also present an integrated system solution for computer- assisted video parsing and content-based video retrieval. The application software package was programmed on Visual C++ development platform.

Welding arc light spectrum in the range of 600nm∼700nm basically composes of continuous spectrum without metal spectrum and argon spectrum. The radiation strength of this continuous spectrum is low and smooth, which is benefit for reducing process, and the response sensitivity of CCD camera is high at this wavelength range. So, choose a suitable imaging spectrum window, use the continuous spectrum of this window to illuminate the welding pool and use CCD camera to sample the pool image. The reflection of arc light from liquid metal pool surface is specular reflection, the reflection of arc light from the workpiece surface is diffuse reflection, which improves the contrast of the welding pool image. This kind of vision image sensing method takes full advantage of the arc light as a benefit factor, and realizes to acquire the comprehensive information of the pool only from a single sensing source. Based on the above principle, this paper develops a visual image sensing system for weld zone of pulsed GTAW. The system as a part of the control system for weld shape can realize simultaneous image sensing of front topside, back topside and bottom side weld pool in a frame. Both the topside and bottom images concentrate on the same target of the CCD camera through the visual sensing light path system. The composite filter technology with low sampling image current is used to overcome the influence of arc light. The high quality and clear images of weld zones are acquired, which supply plenty information to study the dynamic process of pulsed GTAW. In addition, in order to extract the actual size parameters of weld pool, the image sensing system is calibrated.

The target of the research is the image compression and restoration of the texture surface of the buildings for 3-D visualization of cities in this paper. A new method based on texture segmentation and the texture's self-similitude is introduced. The character of the method is that it can remove the interference from the source image, which was obtained in the nature environment, and realize image compression in the same time. The method takes advantage of achieving higher compression ratio than other conventional compression methods, and the algorithm is fast enough to satisfy the 3D reconstruction's need. The restoration image's vision effect is true to nature as a whole. The method is more suitable for the wall texture image compression and restoration of 3D visualization in cities than other traditional compression methods.

In this paper, we present a schema for content-based analysis of broadcast news video. First, we separate commercials from news using audiovisual features. Then, we automatically organize news programs into a content hierarchy at various levels of abstraction via effective integration of video, audio, and text data available from the news programs. Based on these news video structure and content analysis technologies, a TV news video Library is generated, from which users can retrieve definite news story according to their demands.

In on-line signature identification, it is usually difficult to get enough samples. This paper focuses on the work of generating some new signatures according to only a few samples collected advance. In the generating process, we used the knowledge concluded from the observation of lot's of signatures. The knowledge includes the relationship between the writing speed sequences of skilled signatures with different scales, that is to say: the relationship between speed and scale. The knowledge also includes the relationship between writing speed of adjacent points within a signature, which can be consider as the relationship of speed and local surrounding. By modeling the speed sequence of signature as conditional k-th step Markov chain, the relationship of speed and local surrounding is described as the transition probabilities of the Markov chain. The results of the generated signatures show that the method we proposed is useful. The results also prove that the relations we concluded are steady and can be used directly to signature identification.

Each flight experiment needs a lot of labor power and costs much money. Therefore several simulation experiments must be conducted on the ground beforehand. The simulation experiment of a flight above the sea is more complicated than the simulation experiment of TAN (Terrain aided Navigation), since even for a fixed sea region, height data of sea wave are time-variant, so it is impossible to perform a simulation experiment directly using the historical flight data obtained in the sea region before. Instead, all the data must be generated by simulation, including the data of sea wave field, the data of air movements, and the instructions of the controlling system and the flight path of the flying machine. Because the radar height meter is one of the inputs of the controlling system, it is necessary to simulate the radar height meter. We give an effective method of simulating radar height meter provided that the simulation data of sea wave field are known. Compared with other methods, our method is simple and easily performed because the amount of data used is small and the arithmetic is convenient.

To simplify the problem of estimating the motion parameters, in this paper, based on the properties of human motion perception we present a feasible approach, which can estimate the motion direction and the motion extension for a linear motion blurred image. Two main parts are developed in our approach. Hough transform is first used to detect the angle (i.e., motion direction) of a given linear motion blurred image, and quadtree method is secondly used to estimate the extent of point spread function of the same image. Many experimental results have confirmed the feasibility of the proposed approach.

A 3D plenoptic function in the form of a mosaic image is presented in this paper. A sequence of plane images and relative cylinder panoramas are sampled original ly by a camera constrained on a line. Then, these original images are arranged in to a mosaic image by abstracting appropriate circles from the images and pasting them together. Pixels in the mosaic image are indexed in three parameters: camera location along the track line and two angles q and f. When rendering a novel view from a certain camera location, pixels in the view are mapped into view-rays of parameters q and f. Then, the values of view-rays are indexed from the mosaic image. Experimental results show that the proposed method is effective in rendering scene.

The generative technique of the dynamic infrared image was discussed in this paper. Because infrared sensor differs from CCD camera in imaging mechanism, it generates the infrared image by incepting the infrared radiation of scene (including target and background). The infrared imaging sensor is affected deeply by the atmospheric radiation, the environmental radiation and the attenuation of atmospheric radiation transfers. Therefore at first in this paper the imaging influence of all kinds of the radiations was analyzed and the calculation formula of radiation was provided, in addition, the passive scene and the active scene were analyzed separately. Then the methods of calculation in the passive scene were provided, and the functions of the scene model, the atmospheric transmission model and the material physical attribute databases were explained. Secondly based on the infrared imaging model, the design idea, the achievable way and the software frame for the simulation software of the infrared image sequence were introduced in SGI workstation. Under the guidance of the idea above, in the third segment of the paper an example of simulative infrared image sequences was presented, which used the sea and sky as background and used the warship as target and used the aircraft as eye point. At last the simulation synthetically was evaluated and the betterment scheme was presented.

In scene matching, it is important that the quality of real- time image and the matching algorithm impact on the matching results. The focus is on the study of the factors influence to scene matching. In this paper, a digital simulation platform is presented, it is used to generate simulation real-time image, and the simulation study is performed with the above requirements. To develop such a platform, many error sources need to be analyzed and modeled first. Then the real-time image selection criteria, which are summarized from the existing real-time images, can be used to selection of real-time image. Finally, plenty of scene matching algorithms have been e4xerted to compare/analyze the simulation results. The experiments show that the reliability of selection criteria can reach 90 percent at least, and if the algorithm is combined with preprocess algorithm, the reliability can reach up to 95 percent. And zoom factor is the most remarkable one in those factors of influencing the matching result. The influence of existing algorithms is discussed in some detail.

Analysis and simulation of smart munitions requires imagery for the munition's sensor to view. The traditional infrared background simulations are always limited in the plane scene studies. A new method is described to synthesize the images in 3D view and with various terrains texture. We develop the random fields model and temperature fields to simulate 3D infrared scenes. Generalized long-correlation (GLC) model, one of random field models, will generate both the 3D terrains skeleton data and the terrains texture in this work. To build the terrain mesh with the random fields, digital elevation models (DEM) are introduced in the paper. And texture mapping technology will perform the task of pasting the texture in the concavo-convex surfaces of the 3D scene. The simulation using random fields model is a very available method to produce 3D infrared scene with great randomicity and reality.

Scene model is one of key components of remote sensing physical models. We introduce a computer scene model (CSM). And the CSM includes the parameterization of 3D landscape scene of a remote sensing pixel, data structure of the scene, methods and procedures for rendering the scene. In the CSM, generation of 3D architecturally realistic scene of a remote sensing pixel is based on vegetation measurement and statistics. Structure and biophysical parameters of plant firstly are measured by hand and by digital photogrammetry on field. Then, the original data are normalized to obtain the feature parameter set of plant, which is used to rebuild single plant in computer. Position of these plants in a pixel falls four interspersing patterns; random, regular, cluster, and transition. The CSM employs two methods to make 3D graphics of a remote sensing pixel. The first rendering approach is based on the L system, which is applicable to plant with small number of big leaf, such as corn. The second one depends on statistics, which is applicable to plant with big number of small leaf, such as tree and grass. The 3D remote sensing pixel scene is used not only to show 3D pictures, but also to compute distribution function of energy by reflected or emitted by the remote sensing pixel. CMS records all polygons, which make up of the 3D scene, in a text file on an array of polygon to provide the scene for next computations.

An evolutionary algorithm is proposed for obtainment of the matching domain blocks of fractal partition in image compression. It makes use of the partitioned iterated function system (IFS) and fractal image. The technique described here utilizes the evolutionary algorithm, which greatly decreases the search space for finding the self-similarities in the given image. Considering the special properties of the problem, some genetic operators are designed and used in combination with the standard operators in order to improve the effectiveness of the evolutionary algorithm. Both theoretical analyses and experiments show that the algorithm is robust and higher compression ratio and image quality can be achieved.

Generally speaking, Route Planning for unmanned aircraft generates an optimal or approved flight path between an initial position and a desired goal in view of various factors. A successful route planner must be able to handle a series of constraints on resultant road. This paper presents a route planning method for unmanned aircraft based on man- machine conversation. Firstly, we will introduce dynamic planning method and the whole process of panning an optimal 3D route according to the given criteria function. After this route is displayed on the computer screen, operator can modify the route to meet his need. Of course, his operation must accord with all constraints, computer will check out his operation by program and update the latest 3D route on the computer screen if the operation is effective. Secondly, we present the implementation of friendly man-machine interactive interface. The experiment attests that the system could plan a 3D route fitting the realistic conditions.