最新刊期
卷 9 ， 期 6 ， 2004
- 摘要：This paper is the literature bibliography on computer graphics in China in 2003. We collected and classified most of important papers in computer graphics field publised in Chinese, selecting 608 references from 12 important Chinese journals published in 2003 and classified these references into different categories according to their contents. Based on the overviewing and analysis work, we found that the number of researchers and developers engaged in computer graphics-related field is increased a lot in the past fews years, and many conferences are held each year, many high-level achievements are made in China. In addition, computer graphics itself develops, deriving some new research topics or directions.2596|187|0更新时间：2024-05-07
- 摘要：Since segmented noisy image using multiscale markov random fields(MRF) still has very small error blocks or single error pixel in the smooth area,this paper proposes an approach associate edge information with multiscale MRF to reduce error classification.Because images are corrupted by the white gaussian noise,edge is difficult to detect successfully.To detect edge,an algorithm based on the stationary wavelet transform and multiplication between neighbour scales is designed. When there is a significant feature at some position,the wavelet coefficients at that position across scales have high correlation.Such correlation can be taken by multiplication between scales,so edge can be located better.Analysis and experiments demonstrate that the proposed edge detection algorithm is effective and fast.Better results are obtained using the segmentation algorithm associate with edge,most error classification among the smooth image area is corrected.The amount of added computation is only brought by the expense of the edge detection procedure.These new edge detection and segmentation algorithm are more fit for strong noisy images.3149|218|0更新时间：2024-05-07
- 摘要：Tagged MR images provide much motion information, which provides advantage for motion reconstruction of left ventricle(LV), but region inhomogeneity, weak edges, artifacts and influence of tag lines can emerge in images, so the segmentation of tagged left ventricle MR images is difficult. At present, it is almost accomplished manually. This paper introduced an automatic segmentation method of tagged left ventricle MR images, which is based on multistage hybrid processing and is composed of three parts. First the left ventricle is located using morphological method, then whose inner and outer contours are initialized using k-mean clustering, templet matching and the myocardium shape restoration based on skeleton. At last, the initial contour lines are evolved using improved level set method to achieve object boundaries. The performance of this new approach is demonstrated using tagged MR images, and experimental results proved the robustness and effectiveness of this method.3114|227|0更新时间：2024-05-07
- 摘要：Image segmentation is an old and difficult problem in digital image processing. A self-adaptive approach of multi-object image segmentation is presented in this paper. In this approach, OSTU method has been improved on, and square error analyse method is used to self-adaptive confirm the best fit object number. Genetic algorithm is used to optimize threshold value. Segmentation result based on the sample image shows that the algorithm achieves good performance in terms of efficiency of segmentation and segment quality. Expreimental results shows that this approach is effective to segment multi-object image.关键词：multi-object image segmentation;self-adaptive;OSTU method;square error analyse;genetic algrithm3057|203|0更新时间：2024-05-07
- 摘要：Level set methods are powerful tool to segment images, but these algorithms have a large computational burden thus are not suitable for real time processing requirement. In this paper, an new accelerating algorithm of level set method is presented which is implemented on the new generation of graphics processing unit(GPU) instead of on CPU. It first introduced how to implement grid computation for algebraic linear operation and finite difference solution of PDE on GPU by fragment program, and then map the level set solver to GPU. Since GPU is a parallel vector processor for streamed data with big bandwidth for data access, and the result data don't need to be transferred from CPU to GPU for data rendering, so the accelerating algorithm is suitable for real time processing and rendering. In this paper a 2D level set solver for image segmentation was tested with comparison of the performance result between fast marching method and the GPU accelerated method. It shows this method can achieve 60 percent quicker.4148|203|0更新时间：2024-05-07
- 摘要：Regarding the inherent drawbacks of linear interpolation (edge blurring or artifact), nonlinear interpolation technique is studied in this paper. Existing new edge-directed interpolation algorithm utilizes the duality between high-resolution covariance and its low-resolution counterpart to realize adaptive nonlinear interpolation and obtains a much better visual quality of interpolated images than traditional linear interpolations. But interpolated images may still suffer from blurring edges or introducing artifacts around edge area. To achieve a better subjective quality, an improved algorithm is proposed in this paper. It makes full use of local relative information in the low-resolution image. Six downsampled neighbors are involved to estimate local covariance characteristics at high resolution. Two important applications of the algorithm are investigated: resolution enhancement of grayscale images and reconstruction of color images. Experimental results demonstrate that the improved algorithm not only decreases the computational complexity but also succeeds in reducing artifacts on the edges and substantially improves the visual quality of the interpolated images over conventional linear interpolation.3203|228|0更新时间：2024-05-07
- 摘要：Image segmentation owns the important status in the structure of image engineering and is of considerable interest in many fields. The algorithm of edge detection based on the agent searching approach and the correlation of neighbor regions is proposed in the paper. There are four species of agents in the proposed algorithm. The four species of agents can respectively compare the correlation of gray level of neighbor regions with the correlation of geometrical features of neighbor regions at 0 degree, 45 degrees, 90 degrees and 135 degrees to judge whether the pixels lie in the edge of the object or not and finally accomplish edge detection. Experimental results of the proposed algorithm are satisfactory. In addition, there is the comparison of experimental results between the proposed algorithm and some algorithms of edge detection in the section of experimental results, such as Prewitt operator, Sobel operator and Canny operator. The comparison of experimental results demonstrates the proposed algorithm has good performance on some aspects.关键词：edge detection;agent;correlation;image segmentation2243|189|0更新时间：2024-05-07
- 摘要：A novel parallel algorithm for distance transformation based on VLSI hardware structure is proposed in this paper. Distance transformation is a global operation on binary image which is widely used in skeletonization, shape matching, object reconstructing, obstacles elusion, etc. Because of the nonlinearity of euclidean distance, it is not convenient to design parallel and fast algorithm for it. Many deformed weighed distance transformation is adopted in practice. In this paper, we propose a parallel algorithm deferent from the traditional Quasi-Euclidean distance transformation. It can be used in traditional IC hardware and DSP. Theoretical analysis and experiments show that the algorithm is simple, fast and the results can approximate the Euclidean distance more exactly. The Voronoi diagram can be got synchronously. Simulation results on PC are also listed in this paper.3578|228|0更新时间：2024-05-07
- 摘要：Motion estimation is a uncertainty problem, which can't be actually solved because of discontinuity, data distortion and random noise of an image if we only start with the algorithm of MAP (maximum a posteriori probability). In this paper, in order to improve the effect of motion estimation, the fundamental idea of fuzzy data fusion and Gibbs distributing have been adopted to change the computation results of Gibbs energy function, and the risk restriction condition of motion field is effectively brought into the local updating process of GNC (graduated non-convexity function). Moreover, a Gibbs energy function based on the discontinuity adaptive Markov model has been established firstly, which can fuse two classes of vectors, one based on feature and the other on gradient under some restriction conditions; Secondly, a Risk Decision Table about the vectors field have been constructed by some experience information, by which each iterative convergence result was supervised and revised so that data fusing can be well realized. In view of the convergence and robustness of the algorithms, the results of fuzzy fusion are obviously better than that of simple Gibbs's estimation.2732|192|0更新时间：2024-05-07
- 摘要：This paper presents a new method for estimating dense and accurate optical flow based on the motion of planar regions. Different from previous approach, textural information are exploited to organize and constrain the interpretation of the motion. Regions with similar texture properties are supposed to be planar patches. This hypothesis is more reasonable and effective than those hypotheses using the brightness smooth region as planar patches. Parametric flow models are estimated in these textually consistent regions in two steps that first compute a coarse fit according to the coarse optical flow. Then this initial fit is refined using a generalization of the standard area-based regression approaches. Partial fit based on the texture information is employed all through the process. Accurate optical flows are computed finally according to the refined parametric motion models. Experimental results on a variety of images sequence indicate that our method produce accurate flow estimates especially in outdoor environment containing rich texture content. Furthermore the incorporation of the textural information through partial fit and image segmentation provides precise localization of motion boundaries and makes this method overmatch the previous brightness smooth hypothesis based approaches.关键词：optical flow;segmentation;texture constraint;parameterized flow models;partial fit2439|196|0更新时间：2024-05-07
- 摘要：H.264 video coding standard uses motion estimation with multiple block sizes to reduce block-matching errors for higher rate-distortion performance. But the more modes are used, the more computation is needed linearly with the number of allowed block sizes. So a fast block motion estimation algorithm named multilevel successive elimination with half-stopping (MSEHS) is provided in this paper. According to the analysis of statistical characteristics about the best modes, multilevel successive elimination algorithm is described to speed up the searching process for the larger block modes firstly, then using the middle results SAD (sum of absolute difference) and motion vector, a half-stopping principle is defined as a threshold to judge whether it is necessary to continue the motion estimation for the smaller block modes. MSEHS algorithm not only reserves the advantage of multiple block modes in motion estimation but also deletes the redundant computation in the searching process. Experimental results show that the proposed technique provides competitive performance compared with full search algorithm, that is, highly reduced computational complexity, very similar image quality and bitrate.2225|185|0更新时间：2024-05-07
- 摘要：To deal with the image definition of the micro manipulating robot under the vision servo control, a new method for the identification of the image definition is put forward, which is based on wavelet packets. As beginning, the features of the microscope system are discussed briefly. Furthermore, the advantage of wavelet packets applied in the analysis of the signal is introduced. With the analysis of the wavelet packets, the image definition eigenvector is constructed, and the linear classifier system for the image definition is established. To work out of the authority vector and threshold of the classifier system, Fisher rule is applied, which reduces the high-degree system to one-degree system. At the end of this paper, the linear classifier system is found to perform well by experiments.2819|229|0更新时间：2024-05-07
- 摘要：Traditionally, the most commonly used sampling lattice in image processing systems is the rectangular sampling lattice. However, when the signal is a circularly band-limited, the minimum sampling density for a hexagonal sampling lattice is 13.4% less than that for a rectangular sampling lattice. This result is one of the most attractive reasons for considering the hexagonal sampling lattice as an alternative to the rectangular lattice. Almost all the input/output devices only support the rectangular sampled digital images, so the transformation between the hexagonal sampling lattice and the rectangular sampling lattice is discussed in this paper, and the hexagonal sampling matrix satisfied Nyquist Sampling Theorem is given. Moreover hexagonal sampled digital images are non-separable signals. It is difficult to process a non-separable signal directly. In this paper, a multiresolution decomposition of hexagonal sampled images based on a separable filter bank is proposed. A multi-scale decomposition structure is obtained with low computation complexity, and the reconstructed image is presented.3040|202|0更新时间：2024-05-07
- 摘要：Stereo matching is one of the classic difficult problems in the computer vision, and its complexity and precision hedge the capability of vision system to reconstruct the 3-D scene. This paper presents a new matching method based on neural network. On the condition that stereo rectification has been performed, the energy function is built on the basis of uniqueness, compatibility and similarity constraints, which reflects the constraint relations of all matching units of the same lines. It is then mapped onto a 2-D neural network for minimization, whose final stable state indicates the possible correspondence of the matching units. The depth map can be acquired through performing the above operation on the all epipolar lines. The algorithm has two traits relative to the traditional approach. 1. Individual pixel point but not scene point or edge line is adopted as matching unit and dense depth map could be obtained directly. 2. The external input of the nodes is not constant again and is the function of gray similarity of correspondent points. The experiments on the synthetic and real images demonstrate the feasibility of our approach.2891|191|0更新时间：2024-05-07
- 摘要：Based on the research results of“feature”by neuroscience, this paper discusses the definition of“spatial feature”and its applications on targets recognition and classification in remote sensing images of high spatial resolution. The research shows spatial feature consists of two aspects: spatial original feature and spatial functional feature. The former is a reflection of the characteristic of constituent elements of spatial entity, and is for recognition of spatial entity (that means“what it is”); the latter is a functional classification indication of spatial entity (that means“what it does”). The paper also illustrates an experiment about airport recognition and classification ground on a remote sensing image of high spatial resolution.2860|350|0更新时间：2024-05-07
- 摘要：The uneven illumination of aerial images will cause color or luminance difference in an image frame. The difference will further affect the production quality of digital aerial images and the further applications of them. This paper analyses the causes of the uneven illumination of aerial images and characters of them, and deeply studies the application of the MASK dodging in the aerial images. Then based on the MASK dodging principle, this paper presents the processing flow and the corresponding processing method aiming to remove the uneven illumination of the digital aerial images. Finally the experiment is given, which shows that the method presented by this paper can be applied in many cases and avoid the disadvantages of methods based on the mathematical model. The experiment also shows that the method can achieve a satisfied effect and remove the uneven illumination for digital aerial images effectively.3535|210|0更新时间：2024-05-07
- 摘要：The analysis of aerial sequential image is of significant importance to estimating the flight path of the unmanned air vehicle. In condition that there is no proiri knowledge on the model and the scenes in the aerial sequential images change continuously, first the images are filtered because of a large amount of noise. Second, the image is segmented into regions by analysis of the gray histogram, then the edge elements covered by the segmented regions are extracted by the improved Canny operator. In order to extract the model effectively, the concept of region feature comparison factor is proposed. Furthermore, the approach based on this concept to extract the suitable model is presented. And, due to robustness and low computational complexity of the matching method based on the hausdorff distance, it is used to resolve the model tracking in each frame of sequential images. Simultaneously, the model is updated in the process of the tracking on account of the model degradation in the aerial sequential images. In addition, on the basis of the projection principle of the camera, it is analyzed how to estimate the flight path of the unmanned air vehicle via the motion of the model in the aerial sequential images. After an experiment using the real aerial sequential images, it is demonstrated that the model extraction and tracking approach is feasible. It provides the foundation for the estimation of flight path of the unmanned air vehicle by use of the aerial image sequence.2958|202|0更新时间：2024-05-07
- 摘要：The curve discretization is one of the fundamental issues in triangulation of planar region and the parameter domain of curved surface. In this paper, the conception of the curve's local feature point is presented. A fast algorithm for the acquisition of the curve's local feature point is proposed, according to the theory of parallel. Through optimizing the local feature point, the needless local feature point is omitted, and the feature point set requisite is obtained. Accordingly the curve discretization is realized, making use of the curve's local feature point. In the course of curve discretization, the quality influence of the succeeding triangulation, resulted by the tolerance of the discretization precision and the length of the approximation polygon edge, is fully considered in this method. If the approximation precision is high, the curve surface crack can be avoided. If the edge of the approximation polygon is too short, too dense triangle mesh will be created, which will result in long computation time and more memory space, or the long and narrow triangle mesh will be produced, which will make the quality of triangle mesh worse. Experiments indicate that the polygon consisting of the feature points can perfectly approximate the curve, and the algorithm is highly efficient.3432|199|0更新时间：2024-05-07
- 摘要：The estimation of fractal dimension is essential in fractal-based image segmentation. The most popular estimation algorithm is based on box-counting. However, the regular partition and counting methods used in this scheme produces less accurate results. Though morphological fractal estimation is more accurate, the traditional method is very time consuming. In this paper, a new morphological fractal dimension estimation algorithm based on variable structure elements (SE) is proposed. The digitized gray level image is treated as a three-dimensional surface, which is measured under different scales by performing dilations by a series of structuring elements with different sizes on it. And the fractal dimension of it can be estimated from the power law followed by the metrics of the surface and the size of the structuring elements. By properly choosing the structure elements and constructing iterative dilations, the new method achieves better accuracy as well as efficiency. Comparative experiments on both synthetic textures and natural textures show that the proposed approach gives better results than five other commonly used estimation methods. Finally, the estimated fractal dimension and local average gray level are used as characters to segment remote sensing images, comparing with other fractal-based methods, it provides more meaningful segmentation. All these satisfied experimental results demonstrate that the proposed estimation can be successfully applied to texture segmentation.4029|226|0更新时间：2024-05-07
- Postal code：100190
- Tel：010-58887035/58887030/58887418 Email：jig@aircas.ac.cn
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备05080539号-4
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0