摘要:This paper is the tenth of the literature bibliography on computer graphics engineering in China in 2004. We collected and classified most of important papers in computer graphics field published in Chinese, selecting references from 11 important Chinese journals published in 2004 and classified these references into different categories according to their contents. Based on the overview and analysis work,we found that the number of researchers and developers engaged in computer graphics-related field increased a lot in the past 10 years,and many conferences are held each year,many high-level achievements are made in China. In addition, computer graphics develops itself, deriving out some new research topics or directions even cross disciplies.
摘要:Automatic Target Recognition in SAR imagery becomes popular in recent years. The typical Automatic Target Recognition system consists of three stages: detection, discrimination and classification. Detection, whose role is to find regions in SAR imagery that contains potential targets, will inevitably produce false alarms. The false alarms are then further reduced by the following stage, discrimination. These two stages together are called prescreening, which are very important in the whole ATR system. If they act highly efficient, i.e., they can reject almost all the background clutter, the computational cost in the process of classification will be greatly reduced. At present, there are three methods in the field of automatic target detection: CFAR, multi-resolution model, detection methods based on image phases. There are also quite a few methods for target discrimination. In this paper we present an overview on the algorithms and their results for automatic target detection and discrimination in SAR imagery. The research trends of these fields are also given at the end of the paper.
摘要:An elliptical basis function (EBF) network is proposed in this study for the classification of remotely sensed images. Though similar in structure, the EBF network differs from the well-known radial basis function (RBF) network by incorporating full covariance matrices and uses the expectation-maximization (EM) algorithm to estimate the basis functions. Since remotely sensed data often take on mixture-density distributions in the feature space, the proposed network not only possesses the advantage of the RBF mechanism but also utilizes the EM algorithm to compute the maximum likelihood estimates of the mean vectors and covariance matrices of a Gaussian mixture distribution in the training phase, which leas to more reasonable classification. Experimental results show that, compared to RBF network, the EM-based EBF network is more accurate and simpler in structure.
摘要:In recent years, the Bayesian network has been used in many study fields as a data-mining tool, but so far it is seldom used to process remote sensing data. In this paper, we introduce the algorithm about constructing Bayesian network classifier for remote sensing data based on the conditional mutual information test of different bands. The technical procedures of change detection with remote sensing data using Bayesian network are also presented, and the multi temporal Landsat TM data of Beijing acquired in 1994 and 2003 are taken as an example and performed with change detection. The change detection results show that from the year 1994 to 2003, 26.52% farmland of study area had been changed to urban land, 4.68% greenland was increased. The Directed Acyclic Graph (DAG) of Bayesian network describes the mutual information of multi-characteristic data, which synthesized the prior probability and sample information. The study results suggest that Bayesian network will be a newly effective approach for remote sensing data change detection.
摘要:Blind source separation is a promising technique for signal processing, which has such features as blind information processing and waveform restoral. In order to keep the data secret reliably, the applicable principle of BSS in noise removal was extended to form a kind of new method for data secrecy based on the strategy of complete covering, by the introduction of virtual observation. This new method is not only feasible, but of very high security. The results of experiments verified that this new method for data secrecy was simple and effective and had many charming features. For example, it was very simple to use, and has considerable flexibility to implement. Therefore, this method can be used for keeping secrecy of many kinds of different data such as text, speech and image. All of these implied its great potential in information security field.
摘要:In order to improve the accuracy of prospecting and efficiency of oil extraction, it is necessary to understand the geological construction deeply. Therefore, the reconstruction of fault surface models is highly important. A method for fault surface models reconstruction is proposed in this paper. Reproducing kernel developed in different disciplines has become an important tool in functional approximation. By combining the reproducing kernel and neural network, a new kind of neural networks, i. e. the reproducing kernel neural network (RKNN) has been initiated. Besides, learning of the network is converted into seeking the solution of the linear equations system. It is essential to consider the sparse solution so as to construct a simple model with sufficient accuracy and represent the system behavior. The sparse solution is an approximating solution with a large part of components as zero. Although the over all error is small, errors of some points may be very big. The error correction of the sparse solution is also discussed. The reconstruction of fault surface models based on the reproducing kernel neural network is implemented in Daqing, and the experimental results show that the reconstructed fault surfaces based on the method presented in this paper is more suitable for the geological situation in Daqing compared with traditional method.
摘要:The iris image with bad quality will be rejected by the automatic iris recognition system, which will result in the failure of identification or verification. Therefore it is necessary to introduce the quality assessment process of iris images in the phase of image acquisition, which aims to pick out the iris image with high quality from the video sequence captured by the image acquisition device. This paper proposes a fast localization method based on connective region analysis to find out the position of the pupil in an iris image, and then proposes a new iris image quality assessment algorithm which can be divided into two steps. In the first step, the high frequency power of right and left part of the pupil is calculated to find out the undefocused images. And in the second step, the gradient information of the upper region of the pupil is computed to distinguish between the occluded images and unoccluded images. The experimental results indicate that the new algorithm is effective to pick out the iris images with high quality.
摘要:In this paper, a new algorithm based on pattern-list analysis is introduced for page segmentation and classification of document images with irregular-shaped halftone regions embedded in the text regions. This algorithm is composed of three steps. The first step, all the black pixels are extracted by the bounding-boxes and are stored in a linked rectangle-list. The second step, all connected rectangles are grouped to form patterns and pattern-list. At last, the page images are classified into text regions and halftone regions according to their the statistical features. After above three steps, still uncertain patterns are further classified by the type of contextual patterns. Experimental results show the fastness of the proposed algorithm in segmenting text and halftone regions and its excellent performance for complex document images.
关键词:rectangle-list;pattern-list;pattern context;page segmentation and classification
摘要:To improve the training speed performance of large-scale support vector machine(SVM), a fast algorithm is proposed in this paper by exploiting the geometric distribution of support vector in feature space. A support vector information measure definition based on the nearest inter-classes distance is set up and a sort process is presented. Then a reduced number of sample subspace is extracted for support vector training. In addition, instead of the traditional quadratic programming, multiplicative update is used to solve Lagrange multiplier in optimizing the solution of support vector. The samples of rest are used for cross validating till the algorithm is convergence. Experimental results demonstrate that this method has better performance and has overcome the flaw of standard SVM. This algorithm could greatly reduce the computational load and increase the speed of training, especially in the case of large number of training samples.
摘要:A rotation-invariant generalized roughness vector and an adaptive weighted distance are applied in texture retrieval. Texture image is decomposed into amplitude spectra and frequency spectra based on Gabor wavelet. The rotation-invariant vector is calculated on the basis of the local energy distributions, Hurst parameters and direction differences. Then the adaptive weighted distance is used to measure the similarity. In order to demonstrate the efficiency of this method for the general texture images, the experiments are carried on the rotation texture database and the general texture database respectively. The proposed approach performed well in both of the experiments.
摘要:Multi-sensor image registration is important for the spatial data processing, which especially faces the isomerism problems. In multi-modal image registration, the parameters of the transformation between the two images could be estimated by the corresponding pairs of control points. Facing with the difficulties of multi-modal images, this registration algorithm adopts the vision characters and multi-resolution based feature matching to carry out the new coarse-to-fine registration. It is composed of two steps. The first step is to set up wavelet-based image pyramid to decompose images into extract base contours and detail contours. In this step, an extended contour searching algorithm is developed to settle the image noises. The second step is to design a minimum distance classification based multi-resolution shape matching algorithm on the Fourier curvature representation of the chain code contour. Transformation parameters are estimated based on the final matched control-point pairs which are the centers of gravity of the closed contours. The multi-sensor Landsat TM imagery and infrared imagery have been used as experimental data to compare the algorithm with the classical contour-based registration. Experimental results show that this registration is superior to the classical ones.
摘要:Dynamic bandwidth allocation is needed to guarantee quality of service , while video traffic prediction plays an important role in dynamic bandwidth allocation. In this paper, both autocorrelation and Hurst parameter of self-similarity are illustrated to show that traffic character of original frame sequence can be represented by traffic on the time scale of GOP. A new variable step size adaptive algorithm is also presented based on the fixed step size adaptive algorithm, which is used to predict MPEG4 video traffic on large time scale of GOP. And VSSA algorithm can improve performance of prediction significantly according to a number of simulations.
关键词:MPEG4 video traffic prediction;GOP sequence;Hurst parameter;adaptive algorithm
摘要:This paper presents the issue of QP mapping in H.264 or AVS to MPEG-2 transcoding process. The theoretic QP mapping curve is deduced by analysis of H.264/AVS and MPEG-2 quantization schemes. Then, the mismatching between the theoretic curve and practical application is found. Based on the requirements of different applications, such as high fidelity video application, video transmission on limited bandwidth networks and video on demand with adaptive quality adjustment, a series of QP mapping curves is obtained by experiments and analysis. These curves are customized for predefined parameters, and they can be easily modified to meet specific applications.
摘要:Creating polygon is a normal arithmetic through turn left or right algorithm. According to the rule, polygons can be created by clockwise or anti-clockwise. But it may make some repeated and invalid polygons. The purpose of this paper is to introduce a method to judge polygon search direction and creat every polygon by only left algorithm. The method presented in this paper is to judge the polygons search direction to according to the angle changing between adjacent lines. Experimental results show that it is an efficient method to create polygons automatically and avoid invalid polygons.
摘要:There are now many platform software for WebGIS, but none provided a perfect solution to distributed processing. On the base of analyzing the existing software technology of WebGIS, we explored the solution for distributed WebGIS and developed a platform for WebGIS using Java. With this platform, a city spatial information service system based on SIG was established, which can access and process the distributed spatial data.