Search

Speckle suppression is an important step in syntetic aperture radar (SAR) image processing. This paper addresses the problem of reducing speckle in SAR images by employing stack filters. Median filters have been used to successfully reduce non-Gaussian noises and impulse noises. Stack filters are a very large class of nonlinear filters that possess the threshold decomposition and the stack property. In this paper, stacking median filters are introduced. An algorithm for SAR speckle suppression based on stacking median filters is proposed. Threshold selection is also discussed. Experiments are performed using SAR images at different threshold differences to show the proposed algorithm's effectiveness.

This paper proposes a new method to denoise images corrupted by Poisson noise. Poisson noise is signal-dependent, and consequently, separating signals from noise is a very difficult task. In most current Poisson noise reduction algorithms, noise signal are pre-processed to approximate Gaussian noise, and then denoised by a conventional Gaussian denoising algorithm. In this paper, we propose to use adaptive basis functions derived from the data using modified ICA (Independent Component Analysis), and a maximum likelihood shrinkage algorithm based on the property of Poisson noise. This modified ICA method is based on a denoising method called "Sparse Code Shrinkage (SCS)" and wavelet-domain denoising. In denoising procedure of ICA-domain, the shrinkage function is determined by the property of Poisson noise that adapts to the intensity of signal. The performance of the proposed algorithm is validated with simulated data experiments, and the results demonstrate that the algorithm greatly improves the denoising performance in images contaminated by Poisson noise.

Single sensor night vision system has been developed for many years, but it still has many intrinsic limitations. For example, in a single channel imaging system the image contrast will be too low to distinguish different targets. As a new technique, multi-sensor image fusion color night vision is developed. This system can make up a single sensor's limitation and improve the system's performance. Depending on Sharmon's formula of the information theory, this paper describes the formula of the fusion image's amount of information and proves that multi-sensor image fusion technique is good at increasing amount of information. Then according to the information theory, the LLLCCD and IRFPA image fusion system is developed. The system's block diagram is given and described in detail in this paper. And the improved fusion algorithm is adopted which is especially good at this fusion system. At last the impersonal judgment of fusion effect is employed to analysis this system's performance and the practical image is given to show this system's good effect.

In this paper, we propose content-based video retrieval, which is a kind of retrieval by its semantical contents. Because video data is composed of multimodal information streams such as video, auditory and textual streams, we describe a strategy of using multimodal analysis for automatic parsing sports video. The paper first defines the basic structure of sports video database system, and then introduces a new approach that integrates visual stream analysis, speech recognition, speech signal processing and text extraction to realize video retrieval. The experimental results for TV sports video of football games indicate that the multimodal analysis is effective for video retrieval by quickly browsing tree-like video clips or inputting keywords within predefined domain.

In this paper an object recognition method based genetic algorithm was presented. One of the mostdifficult problems in object recognition is to correctly decide whether two dissimilar images are originated fromdifferent items, or belong to the same object but viewed from different camera positions. Invariant-object recognition isto identify an object independently of its position (translated or rotated) and size (larger or smaller). It was found thatimages of objects observed from two different viewpoints can be approximately related by affine transformation if thecamera is placed sufficiently far away. This paper proposes an affine and projective invariant ship object recognitionapproach and employes genetic algorithm (GA) to find the appropriate affine transformation parameters to implementthe matching process between the reference object and the scene object. Experiment results show good performances of the proposed method. Keywords Object recognition, Genetic algorithm, Affine transformation

This paper describes the problem of labeling and completing an imperfect line drawing. Almost all of techniques about shape from contour require a perfect line drawing as their input. Experience indicates, however, that lines are often missing. We exhaustively studied all possible configurations of the imperfect line drawing and found that it is closely related to L-junction. W and Y-junctions in the imperfect line drawings are the same as they are in a perfect line drawing, L-junctions, however, are different because they could be degenerated from W or Y ones. In this paper, it is shown that the number of L-junction in the imperfect line drawing is 36. The 36 L-junctions and their corresponding W or Y ones from which they are degenerated are illustrated. We show how to utilize the line labeling and junction to predict and complete missing line.

In this paper, a system for the segmentation and classification of the skewed document images with graph regions is proposed. In this system, the skewed angles of the document images are detected with a novel algorithm based on the morphological operation of Hit-or-Miss and the Hierarchical Hough transformation. To make the system valid for document images with graph regions, we proposed to introduce a middle point cut process to the traditional recursive X-Y cuts (RXYC) segmentation algorithm so that the graph regions can be approximated with a lot of small rectangles. The segmented regions are classified by two features of BWR and CC, which represent respectively the black to white pixel ratio and the cross-correlation between pixels of the sub-blocks. Experimental results have proved the fastness and the reliability of the system proposed in this paper.

In this paper, we present a new method by using 2-D discrete multiwavelet transform in image denoising. The developments in wavelet theory have given rise to the wavelet thresholding method, for extracting a signal from noisy data. The method of signal denoising via wavelet thresholding was popularized. Multiwavelets have recently been introduced and they offer simultaneous orthogonality, symmetry and short support. This property makes multiwavelets more suitable for various image processing applications, especially denoising. It is based on thresholding of multiwavelet coefficients arising from the standard scalar orthogonal wavelet transform. It takes into account the covariance structure of the transform. Denoising is images via thresholding of the multiwavelet coefficients result from preprocessing and the discrete multiwavelet transform can be carried out by threating the output in this paper. The form of the threshold is carefully formulated and is the key to the excellent results obtained in the extensive numerical simulations of image denoising. The performances of multiwavelets are compared with those of scalar wavelets. Simulations reveal that multiwavelet based image denoising schemes outperform wavelet based method both subjectively and objectively.

With the development of non-contact measurement in the close range photogrammetry, the use of the slide projector becomes familiar and frequent. In order to take full advantage of the slide projector in the procedures of diverse photogrammetric measurements, the slide projector needs to be calibrated at first. Namely, the intrinsic paramters of the slide projector have to be calibrated in advance. In this paper a flexible technique is proposed to calibrate the slide projector easily. The technique only requires an ordinary slide projector, a camera and a planar grid. The algorithm with 2D direct linear transformation (2D-DLT) and collinear equations is used to calibrate the slide projector. The operation method in detail and the algorithm are addressed systematically and entirely. First, the restricting condition among 2D-DLT parameters is worked out using the correspondence of 2D-DLT and collinear equation. Then, the decomposition of initial values of the slide projector intrinsic and extrinsic parameters using 2D-DLT is deduced. Finally, the slide projector calibration parameters are worked out by the whole adjustment. The feasibility and the exactness of the slide projector calibration technique put forward in this paper are verified by the results of real data.

The objective of this paper is to develop a rapid, objective, and easy method for recognizing wheat varieties, which is important for breeding, milling and marketing. The method can be used in place of the existing procedures to remove subjectivity from wheat variety recognition. In contrast to previous work, most of which has focused on wheat morphological characteristics, the features utilized in this paper are based mainly on kernel color. Varietal classification is performed by using Support Vector Machines (SVMs) method. More than 96% correct recognition rates are achieved with bulk samples involving 16 varieties representing a wide range of wheat varieties, wheat class, and kernel types. The proportion of single wheat kernels correctly recognized ranges from 87% to 93%. The results were encouraging since the method proposed here can be easily conducted in routine inspection.

This paper proposes an approach based on using lifting scheme to construct integer wavelet transform whosepurpose is to realize the lossless compression of images. Then researches on application of medical image, software simulation of corresponding algorithm and experiment result are presented in this paper. Experiment shows that thismethod could improve the compression ratio and resolution.Keywords : lifting scheme(LS), integer wavelet transform(IWT), lossless compression 1. INTRODUCTION With the rapid development of the Internet technique, telemedicine has come to our lives. In fact doctors would not find out the symptom, if there were any little information of medical images lost in the transmission. Therefore it becomes urgent to use lossless compression in medical images, in order to ensure the high quality of transmission andstorage. The study has investigated in integer wavelet which is based on lifting scheme, and realize the algorithm of the lossless wavelet transform, which adopts the idea that use integer to integer replacement of adding and subtracting.Compared with other methods, it doesn't have to quantify the transformed data but still could compress the losslessimages completely. We can transmit the medical images with high resolution. It shows that this method has great

This paper addresses two problems of a ship handling simulator. Firstly, 360 scene generation, especially 3D dynamic sea wave modeling, is described. Secondly, a multi-computer complementation of ship handling simulator. This paper also gives the experimental results of the proposed ship handling simulator.

The linear trajectory is one of major performance for industrial robot. A vision-based robots' linear trajectory measure system is introduced in this paper using a structure light and a special measure track. The three inflexions of the optical strip imaging at the V shape track are used to compute the pose between the sensor frame and the track frame, then the linear trajectory of robot can be computed. The emphasis of this paper is the image processing. At this paper, the process of the image processing method for this system will be described at first, then the key methods include image segmentation and line fitting will be discussed, at last the experiment results will be given.

These fusion methods such as IHS transform, Brovey transform and principal components transform could merged two optical image data of different resolutions - a high spatial resolution panchromatic image and a low spatial resolution multi-spectral image. But these fusion methods required the spectral range of the high spatial resolution panchromatic image equals or approximates to the spectral range covered with the multi-spectral image. This paper brings forward a new fusion method called FM that could merge two optical image data of different spectral range. This paper proposed its algorithm, firstly to filter on the panchromatic image, then to merge the remote sensing data applying algebra ratio. The fused production is more excellent at spectral preservation.

Feature extraction is an important step before object detection and pattern recognition are conducted. In this paper, we endeavor to find a new way of feature extraction. As well known, the fundamental of computerized tomography (CT) is image reconstruction from projections. And a famous image reconstruction algorithm is filtered back-projection. Since the filtered projections can reconstruct the original object, it can be inferred that they can also represent the object. One of the contribution of this paper is that the idea image reconstruction from projections is adopted. To reduce the pattern dimensionality without loss of the ability of characterizing the object, the fractal dimensions of all the filtered projections are computed. These fractal dimensions form a feature vector from which pattern recognition can be done easily. Preliminary results have demonstrated that the proposed approach is a promising method for feature extraction.

This paper presents a wavelet-domain Hidden Markov Tree(HMT)-based color image superresolution algorithm using multi-channel data fusion. Because there exists correlations among the three channels of a RGB color image, a channel by channel superresolution method almost certain leads to color distortion. In order to solve this problem, first the low-resolution color image is converted into a gray-scale image using the spatially-adaptive approach presented in this paper and the resulting gray-scale image must reflect the human perception of edges in the color image; then by superresolving this gray-scale image, a high-resolution image is obtained; finally, wavelet-domain HMT-based image superresolutions are performed for the three channels of the low-resolution color image using the same posterior state probabilities, which reflect the hidden states of the wavelet coefficients of the high-resolution gray-scale image obtained before, and thus the resulting high-resolution color image is what we desired. Becasue the correlations among the three channels of a RGB color image are considered, there are no color distortions in the reconstructed high-resolution image. Experimental results show that the reconstructed color images have high PSNR and are of high visual quality.

on the basis of analysis for degradation model of image due to uniform Liner motion, this paper deduces restorationmodel of image blurred by uniform Liner motion, and constitutes a kind of cost function of restoration By analyzing therule of cost function, the paper proposes an algorithm estimating motion parameter The results show that the algorithmis very efficient to the restoration of image bluned by uniform Liner motion when velocity on uniform Liner motion isn'tvery large.Key words: Blind Restoration, Blurred, Motion image, Uniform liner Motion

A new architecture for HIPAS (Hyperspectral Image Processing and Analysis System V2.0) was introduced in this paper which was modified and improved based on the first version of HIPAS V1.0. The comprehensive hyperspectral image analyzing system has been developed under VC++6.0 integrated development environment (IDE) and obtained perfect runtime efficiency and stability. The base architecture was specially designed and implemented to meet the requirements for the rapid preprocessing of imaging spectrometer data and easy prototyping of algorithms. Based on the modularized and object oriented software engineering construction, the architecture is compatible for other UNIX platforms with little modification. The most important components of HIPAS were presented in this paper including tools for input/output, preprocessing, data visualization, information extraction, conventional image analysis, advanced tools, and integrated interface to connect with general spectral databases. Some new methodologies for data analysis and processing were realized and applied to reach some valuable results based on the architecture including mineral identification, agriculture investigation, urban mapping etc. With an open storage architecture, HIPAS is entirely compatible with some advanced special commercial software such as ENVI and ERDAS and even the common image processing system Photoshop. At last, a strict and careful software test was carried out and the results were also analyzed and discussed.

In this paper a FLIR image segmentation algorithm based on genetic algorithm and fuzzy set theory was presented. Image processing has to deal with many ambigious situations. Fuzzy set theory is a useful mathematical tool for handling the ambiguity or uncertainty. A fuzzy entropy is a functional on fuzzy sets that becomes smaller when the sharpness of its argument fuzzy set is improved. The paper defined different member function for the object and background of the image to transform the image into fuzzy domain and chose Z-function and S-function as the membership functions for the object and background of the image respectively and threshold the image into the object and background by maximizing the fuzzy entropy. The procedure for finding combination of a, b and c is implemented by genetic algorithm with appropriate coding method to avoid useless chromosomes. The experiment results show that our proposed method gives better performance than other general methods with good real-time by using genetic algorithm.

Morphological pattern spectrum is a useful shape description tool for quantifying the geometric shape feature of both binary and gray images. In this paper, a general frame of pattern spectrum is developed for both continuous-scale and discrete-scale based on the efficient and reduced redundancy multiscale image representation. A discussion for the basic properties of generalized pattern spectrum is presented in the paper. Algorithms for shape recognition and shape classification using continuous and discrete-scale pattern of images are proposed in Euclidian space.

Image watermarking has become a popular technique for authentication and copyright protection with the development of Internet and computer. However, current image watermarking approaches especially blind techniques are not strongly robust with respect to attacks or combinations of several attacks. In this paper a new intelligent second generation blind image watermarking technique is proposed, which adopts independent component analysis (ICA) for watermarking process. The characteristics of the human visual system (HVS) are incorporated into the watermark embedding, so that the watermark can be adaptive to the protected image. The edge of original image which extracted by Sobel operator is used as watermark in this paper. The watermark is rearranged by chaotic before watermark embedding in order to enhance the robustness of watermarking and the embedding process can be performed in any image domain, including spatial and transform domain. Watermark can be extracted correctly not merely be detected without any information about the original image and original watermark, and the accuracy of watermark extraction depends on the statistical independence between the original image, original watermark and the key. This proposed intelligent system can also extract multiple watermarks embedded in the test image one by one. Experimental results demonstrate that the proposed intelligent second generation watermarking technique based on ICA is robust with respect to attacks produced by popular watermark test software - Stirmark, including rotation, scaling, translation, skew, cropping, filtering, image compression, and combined attacks.

In this paper, a new approach for automatic creating boundary-representation of planar objects from single line drawings depicting the objects was proposed. Compared with other reconstruction methods that use information a bout points, for computer visiona nd CAD system, it was more convenient to use information about lines to interpret line drawings of polyhedral objects. The coordinates relations between a three-dimensional line and its perspective projection in two-dimension plane are given. Another three coordinates relations are also provided, namely, the coordinates relations when two lines are intersecting, parallel, cross, perpendicular, relations when a line is parallel with or belongs to a plane, and relations when two planes parallel each other. On the basis of these coordinates relations, some new constraints about lines and planes are proposed and a linear system is established. The 3D information of planar objects can be obtained by solving the linear system. In the interpretation of line drawing of a planar object, it is proved that there are at least four degrees of freedom. Some examples are given.

An endoscopic image retrieval system based on color clustering method has been presented in this paper. To reduce the color sensitivity to noise and the color histogram dimension and to enhance the accuracy and speed of retrieval, HSV color space is employed in this system and each Hue, Saturation and Value (Intensity) component have been quantified into 6, 8, 4 levels, respectively, because endoscopic images generally contain only a few of dominant colors, such as red, yellow or purple and HSV color space is most approximate to human perception, so the whole HSV space is divided into 192 (6x8x4) subspaces and each subspace is clustered as an index. So a color histogram with 256-dimension is used as indexing and the histogram similarity measure is also given at the same time. The algorithm has been successfully adopted by our endoscopic image retrieval system and the experiment with a database about 1000 clinical endoscopic images has demonstrated its effectiveness and rationality.

In this paper some properties of Foley-Sammon optimal discriminant vector (FSODV), by contrast with uncorrelated optimal discriminant vector (UODV), are discussed. Firstly the Fisher ratio of every FSODV must be not less than that of corresponding UODV and consequently sole FSODV will be superior to corresponding UODV. Secondly the correlation between feature components extracted by FSODV is an important factor. If high correlation is available between most of the feature components, the classification performance of FSODV will be remarkably inferior to UODV. However, if most of the feature components are only little correlative to each other, FSODV is comparative to UODV in classification.

One important step in map vectorization is separation of geograhical information in a cartographic map. Such a fundamental task is not only time-consuming but also laborious. To get an overall understanding of this task and explain why it's hard to achieve a good result, a summary about the characteristics of color map segmentation is presented first in this paper. Then a new algorithm, featuring practicability and adaptability to the variation in brightness and contrast of images, is proposed directly based on RGB color model. Meanwhile, structure features are also used to improve the separation result by connecting broken lines. Experiments on many real map images prove that the algorithm realizes automatic segmentation of color maps and produces a fine result which can be used in automatic data collection without much further processing.© (2003) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Because of the limit of finite missile time response and the field of view (FOV) of imaging infrared (IR) seeker during the endgame of an intercept, the image of the fighter grows larger gradually and finally will overflow the FOV as the missle approaches the fighter. It then results in losing control of the seeker and affecting the precision of burst control of imaging IR fuze based upon the guidance integrated fuzing (GIF) technology. The aim of the research presented in this paper is to decrease the blind range of imaging IR seeker and improve the precision of aim-point parameters through pose recognition. On the basis of the moving characters of missle and fighter during high-speed encounter and the high correlation of frame to frame in image sequence obtained b imaging IR seeker, a novel method of fighter axis pose recognition and aim-point estimation is proposed. Within this methodology, the spatial pose of fighter axis is recognized before the image overflow the FOV, and then the tracking mode of seeker is transformed from general tracking mode to partial image tracking mode in right time. During partial image tracking, the seeker is controlled to keep the partial image track point in the FOV, then the aim-point parameters can be calculated accurately by utilizing the fighter axis pose, parameters of track point and the relative distance of track point and aim-point.

A method for automatic segmentation and detection of small target in the earth to sky background was presented in this paper. When small target in far distance some objects on ground came upon to the scene. This always resulted in false detection in automatic detecting systems if we segmented small object by only using its illumination. In order to remove the big block and disconnected part of a gradient image was made because the gradient scale of the small target and background objects may be comparatively similar. Then an adaptive threshold method was adopted by using the image means as a threshold for several times to segment objects in the gradient image. After that successive over-relaxation was used to incorporate the disperse regions and the nearby isolated points would connect to the big block. The lowest value of valley (nonzero value) was searched and this value was used as the threshold to make binary image. So the connected clutters could be removed only by counting the number of pixels in the connected regions. Eventually the pipeline target detection algorithm was used to process the sequential images to detect the real small target automatically.

A new concept that is the Discontinuous Frame Difference in image sequences is proposed in this paper, and is applied to the Optical Flow Algorithm. The modified Optical Flow Algorithm overcomes the shortcoming of the traditional Optical Flow Algorithm that cannot detect the small moving target whose moving displacement is less than one pixel between two continuous frames. The infinite norm of the Discontinuous Frame Difference Vector is used to preprocess the image sequences to get rid of most of pixels taht are not the pixels of the moving target. And then the instantaneous velocities of the pixels remained by preprocessing is calculated by the Optical Flow Algorithm. If the pixels in an area have the moving continuity and consistency, a moving object is determined. For the preprocessing is able to get rid of most of pixels, the calculation quantity of the Optical Flow is reduced greatly. But the preprocessing is probably to lose some candidate pixels of the moving target, so the Gray Intensity Analysis is used to find these pixels back again. The Discontinuous Frame Difference Optical Flow Field Algorithm can be composed of the parallel structure system, which can detect different kinds of moving objects with different velocities. The experiment result proves the effectiveness of the method.

The modulus maxima of a signal's wavelet transform on different levels contain important information of the signal, which can be help to construct wavelet coefficients. A fast algorithm based on Hermite interpolation polynomial for reconstructing signal from its wavelet transform maxima is proposed in this paper. An implementation of this algorithm in medical image enhancement is also discussed. Numerical experiments have shown that compared with the Alternating Projection algorithm proposed by Mallat, this reconstruction algorithm is simpler, more efficient, and at the same time keeps high reconstruction Signal to Noise Ratio. When applied to the image contract enhancement, the computing time of this algorithm is much less compared with the one using Mallat's Alternative Projection, and the results are almost the same, so it is a practical fast reconstruction algorithm.

Scene matching navigation based on multisensor image fusion is studied in this paper. Pixel and low feature -level fusion images of optical and IR images are used as real time images matching with optical satellite images as base images, in which linear superposition, nonlinear operators and multiresolution image fusion approaches are adopted to acquire the fused gray and edge strength images. Gray and low feature -level scene matching schemes are also employed to execute the scene matcing simulation experiments on the real flight image data, in which scene matching methods are general CCF and MAD algorithms. The experimental results are given to compare the matching performance when taking different fusion images as real time images under certain matching schemes. The scene matching results based on single sensor images are also given for comparison with the results based on multisensor fusion images.

This paper is focused on the restoration of color remote sensing (including airborne photo). A complete approach is recommended. It propose that two main aspects should be concerned in restoring a remote sensing image, that are restoration of space information, restoration of photometric information. In this proposal, the restoration of space information can be performed by making the modulation transfer function (MTF) as degradation function, in which the MTF is obtained by measuring the edge curve of origin image. The restoration of photometric information can be performed by improved local maximum entropy algorithm. What's more, a valid approach in processing color remote sensing image is recommended. That is splits the color remote sensing image into three monochromatic images which corresponding three visible light bands and synthesizes the three images after being processed separately with psychological color vision restriction. Finally, three novel evaluation variables are obtained based on image restoration to evaluate the image restoration quality in space restoration quality and photometric restoration quality. An evaluation is provided at last.

A new algorithm for segmenting color colonscopic images by fusing color, brightness, spatial distance and texture information is presented in this paper. It makes the fractal dimension (FD) as the measurement for texture feature in images and applies a stochastic clustering algorithm that uses pairwise similarity of elements. The clustering algorithm that is based on a new graph theoretical algorithm for the sampling of cuts in graphs, can obtain the optimal number of clusters automatically. The complexity of our method is lower, and its stochastic nature makes it robust against noise. More than 40 colonscopic images have been used to demonstrate the effectiveness of this new algorithm.

For the traditional method of hyper-plane segmentation, the location of hyper-plane in data space was given by statistical method. In the case of the statistical value of regions is smaller than in the region, the statistical method was not effective. The character of genetic algorithm is global searching optimally. Taken this mathematical advantage the location of Hyper-plane could be located easily. In this paper, EOS/MODIS imagery data is used to test this method. The result is proved that Genetic Algorithms-Hyper-plane is better than MLC method by using same training data.

This paper presents an automatic video based face verification and recognition system by Support Vector Machines (SVMs). Faces as training samples are automatically extracted from input video sequences in real-time by LUT-based Adaboost and are normalized both in geometry and in gray level distribution after facial landmark localization via Simple Direct Appearance Model (SDAM). Two different strategies for multi-class face verification and recognition problems with SVMs, "one-vs-all" and "one-vs-another", are discussed and compared in details. Experiment results over 100 clients are reported to demonstrate the effectiveness of SVM on video sequences.

In this paper, we proposed a new object-based coding algorithm by using wavelet transform to instead of the image encoder algorithm by using FGS in MPEG-4. The new object-based coding algorithm combines motion estimation with object-based 3-D wavelet transform for video coding in order to fully utilize the redundancy in the time domain. The shape-adaptive algorithm based on modifying boundary extension method of lifting scheme. A sequence of VOPs are fed into the motion compensated lifting (MCLIFT) wavelet coder which first decomposes the VOPs temporarily through MCLIFT filter, and then decompresses the VOPs spatially by shape adaptive lifting wavelet transform (SA-TWT). We encode the video and represent the stream as multilayer bit stream. The integrated transport-decoder buffer ensure the video be continuously transmitted. Losing package can be recovered by using re-transmission.

A novel method of optimizing feed-forward neural networks using cascaded genetic algorithm is proposed in this paper. It adopts a hybrid encoding method, which architectures and connection weights vector of neural networks are encoded into binary code and real-value code respectively. The proposed optimizing method includes two cascaded evolutionary procedures in which the first mainly plays the role of fast search in constrained area and the second extends global exploration ability. The proposed method has represented a particular compromise between exploitation and exploration of searching optimized neural networks and enhanced the global search ability while using less computation. The experimental results have shown its good performance.

Scene change detection is an essential step to automatic and content-based video indexing, retrieval and browsing. In this paper, a robust scene change detection and classification approach is presented, which analyzes audio, visual and textual sources and accounts for their inter-relations and coincidence to semantically identify and classify video scenes. Audio analysis focuses on the segmentation of audio stream into four types of semantic data such as silence, speech, music and environmental sound. Further processing on speech segments aims at locating speaker changes. Video analysis partitions visual stream into shots. Text analysis can provide a supplemental source of clues for scene classification and indexing information. We integrate the video and audio analysis results to identify video scenes and use the text information detected by the video OCR technology or derived from transcripts available to refine scene classification. Results from single source segmentation are in some cases suboptimal. By combining visual, aural features adn the accessorial text information, the scence extraction accuracy is enhanced, and more semantic segmentations are developed. Experimental results are proven to rather promising.

Trademarks' retrieval has obtained more and more attention in recent research on content-based management and utilization of image database system. To retrieve the images, the key is to get the shape features. In this paper, a new method for extracting shape features of trademarks is presented. Based on the theory of information, the method uses images' entropy and invariant moments to capture the shape and spatial information of images. The algorithm is easy and the experimental results show its invariability with respect to translation, scale and rotation of objects. What's more, it also have the noise invariance.

This paper discusses the existing three optimal band combination rules of hyperspectral remote sensing images. They are joint entropy, optimal index factor and Sheffield index respectively. Three bands of MODIS images data are combined arbitrarily according to the three rules, so the best three bands combination images of the three rules are acquired. On the basis of this, the three images are all classified in term of maximum likelihood classifier. Also, the influence of each band combination to the classification performance is discussed. The experiment result proves that the best classification performance of the MODIS images based on the three bands combination is the combination image based on optimal index factor.

Recently, SAR interferometry (InSAR) is popular in low-resolution DEM generation and high precision deformation measurement. There are many uncertain errors in InSAR processing that cannot be modeled and eliminated, such as tropospheric delay, phase unwrapping error, baseline error, and propagating noise. As we cannot decrease these errors just by SAR data processing, many other means have been presented to solve the problems. GPS signal and SAR signal are microwave of the same kind, and have the same attribute in propagation, so we can integrate GPS data into InSAR data to eliminate the same propagating errors in the aerosphere. What is more, GPS can observe high precision deformation better than differential InSAR has done. It is valuable to compare the results derived from these two methods. Researaches undertaken by some institutes on integration of GPS and InSAR have shown a prospective future. Therefore, this paper aims to analyze the methods of GPS and InSAR integration and apply them into the case study of city subsidence.

In this paper, an approach for the automatic extraction of linear feature, in particular roads, from digital aerial imagery is proposed. In some literature, knowledge based automatic road extraction were done with geo-referenced imagery which can automatically register old road map to new imagery as knowledge. Whereas the automatic geo-reference and road extraction are processed simultaneously in this approach, which can benefit and depend on each other. The implemented approach is based on Snake model and template matching with the new aerial imagery and old road map. The presented procedure does not need much manual interaction and therefore has the potential to be integrated into an automatic workflow. Potential applications of the approach are manifold, like automatic change detection of road and three-dimensional reconstruct of man-made objects such as road and building wtih some minor modifications.

Anchorperson shot detection is of significance for video shot semantic parsing and indexing clues extraction in content-based news video indexing and retrieval system. This paper presents a model-free anchorperson shot detection scheme based on the graph-theoretical clustering and fuzzy interference. First, a news video is segmented into video shots with any an effective video syntactic parsing algorithm. For each shot, one frame is extracted from the frame sequence as a representative key frame. Then the graph-theoretical clustering algorithm is performed on the key frames to identify the anchorperson frames. The anchorperson frames are further refined based on face detection and fuzzy interference with if-then rules. The proposed scheme achieves a precision of 98.40% and a recall of over 97.69% in the anchorperson shot detection experiment.

Projection Pursuit is applied to explore the potential structures and characters of the multi-dimension data through projecting the high dimensional data set into a low dimensional data space while retaining the information of interest. For the application of hyperspectral images analysis, the detection of small man-made objects is very difficult. But the man-made object can be viewed as anomalies in an unknown environment due to the fact that their spectral is different from those of the large known background. This paper presents a method to detect man-made objects of hyperspectral images based on Projection Pursuit. Also evolutional algorithm was developed in order to find the optimal projection index.

Three-dimensional volume reconstruction has gained great popularity as a powerful technique for the visualization of volume datasets such as those obtained from X-ray, computed tomography, and magnetic resonance imaging in recent years. Local features play important part in the classification process for a variety of medical image analysis, computer-aided diagnosis, and three-dimensional reconstruction and visualization applications. By using high-order local statistic features detected by local block based moments, such as flat, round, elongated shapes, together with the local spectral histogram of textures, to act as classification criteria, a three-dimensional medical reconstruction method is proposed in this paper. A volume splatting algorithm by using the proposed classification method is implemented and relatively high-quality rendering results can be obtained when the proposed method is applied in medical reconstructions.

An unsupervised classification method combining Principal Component Analysis (PCA) and Gaussian Mixture Model for hyperspectral image is proposed in this paper. It is based on the property that lower dimensional linear projections of high dimensional data sets have the tendency to be Gaussian, or a combination of Gaussian distributions as the dimension increases. The spectral dimensionality of the data is first reduced by a PCA linear projection; then the transformed data is modeled by a Gaussian mixture models, the parameters of the model are estimated using the Expectation-Maximimization (EM) algorithm in merge operations and the number of components is automatically selected based on Bayesian Information Criterion (BIC); finally the data after PCA transform is classified according to the mixture model. Applying the method to Push-broom Hyperspectral Imager (PHI) data shows that the method is quite effective without any a prior information.

Reconstruction of 3D objects and scenes based on 2D images is one of the most important topics in computer graphics and computer vision. Traditional visual hull reconstruction methods cannot correctly handle the situation wehre part of the object's concave surface is not reflected on the 2D silhouettes -- for example, the inner surface of a cup cannot be recovered frmo its photo images at any angles. In this paper, a new algorithm is presented to solve this problem, focusing on the passive reconstruction with local application of active methods as a complementation. The basic idea is to first obtain a 3D model from the octree hull reconstruction method, then apply some assistant facility to get the local information of the concave surface, and finally combine the two pieces of information together to obtain a more accurate surface model of the object.

In this paper, a new fault diagnosis approach for the photovoltaic array, which is based on infrared and visible image fusion technique, is proposed. Firstly, the temperature difference and infrared characteristics are analyzed, and then the features of both infrared image and visible image are extracted. By comparing the features of infrared image with those of visible image, the abnormal operating regions covered with something are distinguished. A fuzzy fault diagnosis approach is introduced and implemented for other regions detected in infrared image but none in visible image. Experimental results show that the proposed approach is feasible and effective.

When compared with speech technologies in speech processing, automatic language identification is a relatively new yet difficult problem. In this paper, a language identification algorithm is provided and some experiments are conducted using OGI multi-language telephone speech corpus (OGI-TS). Then experiments results are described. It is shown that GMM-UBM is another efficient method to language identification problems.

This paper present an approach based on improved hybrid elastic model for deformable image registration. This method utilizes the linear spring net model for correspondence, and the thin-plate spline for non-rigid mapping. Compared with the original model, this improved method uses not only the intensity information, but also the weighted gray level histogram and the image gradient for similarity measurement. A multi-resolution strategy is involved to decrease the computing complexity and to increase the precision. Some experiments are performed on both synthetic and segmented medical images.

This paper presents a novel digital image watermarking scheme that is invariant to rotation, scaling, and translation (RST). We embed watermark in the log-polar mappings of the Fourier magnitude spectrum of an original image, and use the phase information of the original image to rectify the watermark positions. The scheme avoids computing inverse log-polar mapping (ILPM) to preserve image quality and avoid exhaustive search to save computation time and reduce false detection rate.