最新刊期
卷 18 ， 期 12 ， 2013
-
Image codec research on the optimum directional selection of wavelet sub-band and compressed sensing
摘要：For the problem of large decoding errors generated by the image compression based on Bayesian compressed sensing with local optimization,a new codec uniting the optimal directional selection of wavelet sub-bands with Bayesian compressed sensing is proposed,which reduces the error of the decoded images and increases the accuracy of the codec. According to the correlation differences between the low-frequency and the high-frequency sub-bands,we use a respective codec for each of them. Low-frequency coefficients contain the images' main information. With the purpose of avoiding local optimization,we apply the correlation of the low-frequency sub-band,introducing the optimal direction selection codec to the low-frequency coefficients,and combining Bayesian compressed sensing based on the wavelet transform to deal with high-frequency coefficients containing details of images. The experimental results show that in the case without increasing the bit rate,the decoded images using the proposed method have good subjective quality,and the PSNR and SSIM of images are improved to some extent.关键词：wavelet sub-band;optimum direction;compressed sensing;sparse representation;image codec2809|340|0更新时间：2024-05-07 - 摘要：To increase the amount of embedded information hidden in digital image, a hiding algorithm based on bit-plane and the information of HVS (human visual system)is proposed. Our proposed method takes both, the position of the image and changes in the pixel gray values into account. First of all,the secret image is stratified by a bit-plane,scrambling in accordance with the improved Arnold transform algorithms.The most significant bit plane is matched with the bit plane of the cover image and then obtained. This reduces the space of the secret sequence,and then in accordance with the masking effect of the human visual system, in the table of embedded depth,the information is embedded. The experimental results show that the algorithm is not ensuring the visual effect,the amount of information which can be embedded is further improved, and the capacity is more than 28%,PSNR greater than 45 dB.Furthermore, the proposed algorithm is more robust.关键词：information hiding;bit-plane;Arnold transform;MSB;HVS2774|252|0更新时间：2024-05-07
- 摘要：Because of the singleness of the rotation factor, fractional Fourier transform (FRFT) is rarely used in real-time video encryption. While at present,the simple chaotic encryption algorithms have many security problems. Consequently,a new real-time video encryption algorithm-chaotic key modulating discrete fractional Fourier transform (DFRFT) rotation factor is proposed. This algorithm combines chaotic encryption with DFRFT. First, we modulate the DFRFT rotation factor by a chaotic key. Second,we use the modulated rotation factor to carry on fractional Fourier transform,so as to complete the encryption operation for video data. The encryption system realizes real-time acquisition,real-time encryption and transmission, as well as real-time reception,real-time decryption and playing at the receiving terminal. The experimental results show that the system has a good effect in the encryption and decryption and can meet our real-time and security requirements. The theoretical analysis on the experiment results indicate that this algorithm is simple and practicable, with a high-level of security. The security of the algorithm is better than simple chaotic encryption algorithm or the simple FRFT encryption algorithm. We can draw a conclusion from results that this algorithm provides a novel mechanism for resolving the dispute between real-time and security.关键词：video data;chaotic key;fractional Fourier transform;rotation factor;real-time transmission2138|282|0更新时间：2024-05-07
- 摘要：Fabric images are easily disturbed by noise and different surface materials. A hybrid approach is proposed to realize effective automatic fabric detection in this paper. Image enhancement and defect detection are combined to solve these problems. Principal component analysis (PCA)is introduced into the similarity evaluation in the nonlocal average filtering algorithm (NLM)to deal with the de-noising process. The GLCM (gray level co-occurrence matrix)texture features are improved by the PCA-NLM (principal component analysis-non local means)algorithm,which increase the class separability between defect areas and non-defect areas. The experiments show that,compared to the non-hybrid methods, the proposed hybrid model can obtain a higher accuracy with seven class fabric defects.3364|373|0更新时间：2024-05-07
- 摘要：In unconstrained environments,face recognition results can be seriously affected by inner and outer factors such as expression,attitude,light conditions and background. In this paper,we mainly study the image alignment based AAM (active appearance model) and local matching approach for face recognition that will be able to enhance the robustness to the change of attitude and expression.AAM can rapidly and accurately locate facial feature points,and then warp the picture into a "standard positive" face model.Several models based on Gabor feature have been proposed for face recognition with very good results on available face databases. In this paper,a methodological improvement on Gabor features is proposed and used to align face data by AAM. We select and weight Gabor jets by entropy measure.Then, we bring a threshold to the Borda count classification,eliminating low score jets produced by the voting and consequently,increasing the face recognition rate. Experiments indicate that combination of weighting Gabor jets features with Borda count thresholding can yield the perfect results on face data aligned by AAM.关键词：face recognition;active appearance model (AAM);Borda count;entropy weighted;gabor feature2605|316|0更新时间：2024-05-07
- 摘要：Existing fingerprint identification algorithms require higher quality fingerprint image and they are not robust for transformed fingerprint recognition. The topology pattern, which is based on the fingerprint topological properties and relationships has different levels of geometric invariance from the complete image to the local image parts and has a higher robustness in recognizing deformed fingerprints. In this paper,two different topology patterns, which are based on the attributions and topological relationship among the fingerprint minutiae and ridges are constructed. The experiment results show that the two constructed topological patterns have strong robustness for the deformed fingerprint image and can be used very well with images of poor quality especially for the slightly transformed fingerprint recognition.关键词：transformed fingerprint recognition;topological pattern;fingerprint minutiae;ridge line;robustness2251|304|0更新时间：2024-05-07
- 摘要：As an important branch of image processing and computer vision,dynamic target detection and tracking is widely applied in military and civilian applications. A new method of target detection and tracking based on fast computation using sparse optical flow is proposed in this paper. Only optical flow vectors of specific pixels which can reflect features of the image are calculated in this method. Furthermore, an image pyramid is combined to detect and track the faster and the larger-scale motions. In this paper,the new method is compared with methods based on dense optical flow and color feature. The comparison results show that the method proposed in this paper has many advantages,such as high calculation efficiency,well dealing with target occlusion,well detecting and tracking fast targets, and so on. Experiments under various conditions are done to validate the effect of this method. Tracking accuracy can reach more than 80% in most cases and the method can also meet the real-time requirement. This indicates that the method is feasible and practical.关键词：sparse optical flow;fast computation;specific pixels;target detection;motion tracking2874|329|0更新时间：2024-05-07
- 摘要：Natural scenes contain a large quantity of image information, but also plenty of text information. These text characters could offer important semantic information. Automatic detection and recognition of text in natural scene images is an important research topic in pattern recognition and image processing. In this paper,we propose an effective method to extract text from scene images. The main idea is as follows:First,an edge detection method is used for coarsely locating the text areas,then we apply a gray based detection scheme to the located areas in order to confirm the found characters. Finally,the noise regions are removed through filtering all the detected regions,and the target text regions are obtained. Experimental results show that our scheme is robust on finding text in natural scene images with respect to different font sizes,styles,colors and orientations.In this way, the text information can be located and extracted accurately.关键词：natural scene image;text localization;edge detection;gray based detection2964|363|0更新时间：2024-05-07
- 摘要：Image segmentation is a main field of application in parallel computing.To achieve the high performance needed,the algorithm needs to make use of the improved computer hardware and parallel computing algorithms. Mean Shift algorithm is a relative classic algorithm in image segmentation field,which needs no prior knowledge and is an unsupervised segmentation process,attracting widespread attention for its good applicability. In this paper, we give two parallel improvement methods of Mean Shift using TBB(threading building block)and CUDA(compute unified device architecture)based on Multi-core and GPU(graphic processing unit) processing. First, the most time-consuming part the Mean Shift iteration in the process of Mean Shift image segmentation, is analyzed,then two parallel improvement methods of the Mean Shift iteration using TBB and CUDA are given, Two parallel methods are compared and analyzed. The experimental results show that,two kinds of parallel methods have achieved preferable acceleration effect,with the increase of the image and bandwidth parameter the speedup of two parallel methods is on the increases,and the speedup based on TBB tends to be equal to the number of CPUs.关键词：Mean Shift;parallel computing;threading building block (TBB);CUDA;image segmentation3899|319|0更新时间：2024-05-07
- 摘要：Foreground detection is a significant step of information acquisition in intelligent surveillance. The task is to segment all the moving objects from complex scenes without any false targets and noise interference. This step is a premise of the following steps: object identification, object tracking and behavioral analysis. Due to non-stationary surveillance scenes, foreground extraction becomes a complex task with many challenges. The performance of foreground detection mainly depends on the background modeling algorithm. In order to solve this problem, an adaptive background modeling approach is proposed. This approach is based on a Gaussian mixture model proposed by Stauffer and Grimson. In their approach, each pixel maintains a Gaussian mixture model constituted by Gaussians. Then each Gaussian mixture model is updated by new observe pixel value. However the strategies of updating have some limits, such as fixed Gaussian number, fixed parameters, and fixed learning rate. The proposed approach optimizes updating strategies so as to break these limits. In this approach, each pixel maintains a dynamic Gaussian mixture model, while the number of Gaussians can be controlled dynamically. Further more, an online EM algorithm is applied to the method for estimating the parameters in Gaussian mixture model. At last, several strategies are proposed to control the learning rate of weights. Experimental results show that the foreground object detection approach has good adaptability to complex environments. The foreground object can be detected effectively and rapidly, and the precision and recall ratio of results demonstrate superiority of the method to some related work.关键词：Gaussian mixture model;foreground detection;background modeling;dynamic control3432|301|0更新时间：2024-05-07
- 摘要：Focusing on the problem of robust object tracking, we propose an object-tracking algorithm based on dictionary learning and practical filter. We represent local image patches of a target object by codes with offline dictionary learning and pose object tracking as a binary classification problem.A learned classifier and a practical filter are used to estimate the tracking result. The proposed method can solve the difficulty caused by illumination variations, complex backgrounds,occlusions and so on.We do the simulations on a variety of challenging sequences. Experiments on different sequences with evaluation of the state-of-the art methods show our algorithm performs favorable performance.关键词：dictionary learning;object tracking;practical filter;classifier3272|349|0更新时间：2024-05-07
- 摘要：To solve the disadvantages that a traditional camera cannot track moving objects in panoramic range, this paper presents an idea about bionic compound eye panoramic detection and a related tracking strategy. On the hardware side, we use multiple sub-eye cameras to detect the whole field of view in real-time, while the center of sub-eye cameras is a large-aperture dominant eye cameras which is installed on the pan tilt and located precisely by a pan tilt controller that receives different codes. On the software side, we present an idea about automatic windowing collection. Multiple layers of windows are collected from the image of the movement area, while non-sport areas have fewer layers windows. Then, we use the method of the Gaussian model to detect the moving object. The detection method is using lateral inhibition in the overlap area to obtain the clear moving object silhouette. In accordance with the above ideas on the hardware and software sides, we build the actual experimental device, and do experiments repeatedly. Compared to the traditional camera's tracking performance, this camera has some characteristics are that the real-time detection and the combination of the windowing acquisition and lateral inhibition, so we obtain a greatly improvement in the field of view and sensitivity.关键词：panoramic detection;background subtraction;motion detection;lateral inhibition2555|243|0更新时间：2024-05-07
- 摘要：Automatic extraction of brain tissue from MRI volume is an important step in the preprocessing of brain analysis. To improve the extraction result, a region of interest based method for automatic brain extraction was proposed. The first step of the proposed method was obtaining the brain region of interest with the brain extraction tool (BET) algorithm. Then a modified hybrid level set model was defined to obtain the true brain boundary. The modified hybrid level set model used a nonlinear speed function which can eliminate the boundary leakage effectively. 18 MRI volumes from internet brain segmentation repository (IBSR) were used in our experiment. The results obtained from our method were very similar with the results using manual extraction and almost achieved the best results on IBSR data compared with other methods. Experiment shows the proposed method is effective and robust.关键词：brain extraction;hybrid level set;region of interest;cerebral MRI;active contour3650|281|0更新时间：2024-05-07
- 摘要：Mesh stitching and fusion is a fundamental operation in a lot of 3D shape editing and modeling applications. For example, it always need to assemble different mechanical apparatuses together in the area of computer-aided industrial design, creating new toys from some existing ones in the area of digital entertainment and reassembling fractured archeological artifacts in the area of cultural relic protection, etc. In general, to blend two under-stitching meshes or fuse several interested sub-parts together, it is widely recognized that the transition surface connecting them should possess the following properties. The transition surface should smoothly combine the underlying meshes in a seamless natural manner for different joining boundaries, whilst the local geometric details should also be preserved as soon as possible in the vicinity of the stitching boundaries. Based on the Hermite interpolation scheme, a new approach of mesh seamless stitching and fusion is presented in this paper, which can be adapted for blending two meshes with the arbitrary distributed boundary point sets. First, their boundary point sets of two under-joining meshes are automatically selected to form the blending region. The two joining boundary curves can thus be interpolated by two quadratic B-spline curves separately. Then, the transition surface can be constructed by Hermite functional blending scheme under the geometric position and the tangential direction limitations of two boundary curves. Finally, the transition region can be created by triangulating its discretely sampled vertices and applying Laplacian smoothing to form the resultant blending mesh. Compared with the traditional mesh fusion methods, owing to interpolating the two boundary curves of two under-fusion meshes by using B-spine curves, our mesh stitching and fusion scheme can be applied to blend the underlying meshes with different types of boundary curves, that is, it is not only adaptable for the meshes with planar boundary curves but also for the meshes with spatial boundary curves. Meanwhile, due to constructing the transition region by Hermite interpolation scheme that can satisfy the position and the tangential continuity constraints of the stitching boundary curves, the generated transition surface can smoothly blend the underlying meshes and reconstruct the local geometric details in the vicinity of the joining boundaries. Moreover, different from representing the blending surface by an implicit function, our explicit Hermite interpolation scheme is both simple and efficient. The experimental results illustrate the effectiveness and the robustness of our poroposed mesh blending approach in many applications, such as the mesh stitching and mesh repair operation for artifact scanned models, the dental crown restoration in the practical dentistry CAD application, and the extended mesh fusion application for combining several parts of different scanned models into a single object. Here, as an extension of mesh stitching and mesh repair, the mesh fusion operation can also be efficiently conducted by Hermite interpolation between every two boundaries of several shapes, which can provide the users in the digital entertainment area or the engineers in the industrial design area a convenient modeling tool to easily create various desired interesting complex 3D shapes.关键词：mesh stitching;mesh repair;mesh fusion;Hermite interpolation;B-spline interpolation6511|596|0更新时间：2024-05-07
- 摘要：Study on computer generation of animations with Chinese traditional styles is of great importance for carrying forward Chinese traditional arts and enhancing the competitiveness of Chinese animations. Propitious clouds are one of the most typical Chinese-style auspicious patterns with a long history in China. However, most of previous research focuses on realistic cloud animations, while little effort has been paid to the cloud animation with traditional Chinese style. In this paper, a cloud model with traditional Chinese style is presented. First, we construct the cloud skeleton curve by use of eclipses, we then arrange a series of primitive circles with different radiuses along with points in the curve and obtain the cloud silhouette by a union of these circles. We construct the animation of silhouettes and cloud shapes by control primitive circles moving along the circumference whose center is on the skeleton curve. We take a dynamic silhouette point as start position and draw a spiral curve along the center of cloud. Experimental results show that the proposed model is able to generate a wide variety of animations of clouds in the traditional Chinese style.关键词：cloud;modeling;computer animation;Chinese traditional style4234|243|0更新时间：2024-05-07
- 摘要：As the impact of natural environment, resource competition, propagation of trees and other factors, the spatial distribution of trees in forest is changing continuously. In this paper, we put forward the method to simulate this dynamic evolution of forest based on the theory of forest ecology. The spatial distribution information of trees is recorded and visualized in a form of two-dimensional distribution diagram, and three important evolution rules of forest: self-thinning, species succession and clustering growth, are simulated with iterated deduction manner. Based on the simulation results above, we realize the visual performance of the procedure of forest dynamic simulation. In the simulation, for the seek of speeding up the process of making intersection test among trees, which is the most time-consuming process in iterated deduction, we propose the improved quadtree algorithm quadtree with extended boundaries to reduce the number of intersection test. Generation the initial distribution information of trees refers to the conception of the foresty spatial data, which can make the initial data of forest to better fit the diameter distribution of trees. The method can be used to provide the dynamic distribution data of forest for further realizing the realistic simulation of dynamic forest scenes. Furthermore, it also has potential value in forestry research and applications. The proposed method can be applied to the areas like film and animation, virtual reality, but also has great potential of research and application of forestry science.关键词：dynamic evolution of forest;visual simulation;spatial distribution of trees;quadtree4170|283|0更新时间：2024-05-07
- 摘要：The existing visual dictionary construction methods need to combine several features into a vector. Then the vectors are clustered to form the dictionary. Those approaches only take the similarity of all the features into consideration but the neglect distinct roles of diverse features on the construction of the visual dictionary. In this paper, a visual dictionary construction method based on Dempster-Shafer (D-S) evidence theory is proposed. D-S evidence theory is applied to fuse different features in their similarities, which is helpful to obtain more accurate visual dictionaries. Two kinds of features are applied in this paper to subdivide the initial visual dictionary based on the Dempster-Shafer evidence theory, and similar features are clustered together better. Compared with the traditional visual dictionary generation method, our proposed method obtains better results. The experimental results on image classification using support vector machine (SVM) and Naive Bayisan (NB) classifiers show that our proposed method outperforms the K-Means based dictionary construction algorithm in terms of accuracy in visual dictionary and image classification.关键词：image classification;visual dictionary;features fusion;Dempster-Shafer evidence theory4189|627|0更新时间：2024-05-07
- 摘要：With the rapid development of the applications of network, enormous multimedia files are emerging every day. Video data is an integration of text, sound, image and other files. Not only being hierarchical, structural and complex, video data is also rich in semantic information. Therefore, extensive attentions have been drawn to how to process video data quickly, extract the video characteristics accurately, and analyze and understand the semantic content deeply. Semantic event detection and analysis could find video information quickly and accurately for users in the vast ocean of video data. It can be applied to the field of video on demand, intelligent monitoring and video mining as well. However, there are still many limitations, such as low recognition rate for multiple moving objects with different characteristics, low accuracy in semantic event detection, difficulties in detecting semantic event correlations, the lack of consistent standards of event semantic description, and so on. The detection and analysis methods of complex event based on matching integration between trajectory and multi-label hypergraphs are proposed. Trajectory and multi-label hypergraphs are constructed for classifying and recognizing the complex events. By matching the trajectory hypergraph and multi-label hypergraph, mapping relationship between trajectory and multiple semantic labels is built to extract the complex semantic events. The recognition of low-level features to high-level semantic is made possible for video events. Compared with other methods, such as event detection method based on graph and multi-label semi-supervised learning methods based on hypergraph, the proposed method has a higher mean average recall rate and mean average accuracy rate in the detection result of the complex event. we propose a new event detection method that is trajectory and multi-label hypergraphs model in this paper. This model is a widely-used detection and analysis of complex semantic-based events method. In their clustering process, their number of vertices and clusters has been more than other graph methods. But, the event detection result is better than others.关键词：hypergraph;trajectory;label multi-label;pairing fusion;spectrum hypergraph cut3848|245|0更新时间：2024-05-07
- 摘要：The single watermark algorithms always merely have a single function. In order to overcome this drawback, a multi-purpose dual watermark algorithm of color images based on the speeded up robust feature(SURF)is proposed in this paper. First, the original image was transformed from RGB to YUV space and the SURF points and vectors of original image are extracted. A 64-dimensional vector is constructed according to the main direction of the SURF point that meets the scale condition, then the vector is split into two 32-dimensional sub-vectors. The robust zero-watermarking sequence is produced by comparing the cosine angles between the sub-vectors and a secret reference vector. Then, the original image is divided into 2×2 sub-blocks and the fragile watermark is produced after XOR operations on the singular value norms of sub-blocks. Lastly, the fragile watermark is embedded into the least significant bit(LSB) in the space domain of the image original. copyright is identified by the bit correct rates(BCR)between the original robust zero-watermarking sequence and the watermark sequence extracted from the tested image. Content authentication is accomplished by judging whether the fragile watermark is consistent with the image least significant bit. The experimental results show that the algorithm not only has good invisibility and high computational efficiency, but also can achieve the dual function of copyright protection and content authentication.关键词：dual watermark;speeded up robust feature (SURF);zero-watermarking;singular value decomposition;norm3740|503|0更新时间：2024-05-07
- 摘要：In non-photorealistic rendering research, people found that many arts' textures or backgrounds are composed of many small elements with repetitive distribution. Many researchers tried to use procedure based approaches to simulate those pattern effects, but those methods could only generate limited kinds of patterns even after a tedious complex and difficult parameters' adjustment. Some researchers considered sample-based methods and have been proved to be very effective in non-photorealistic effects simulation. Vector texture pattern means the pattern which consists of limited types of elements which are arranged with certain rules or features. Those patterns are very common in our daily life, both from natural objects and manmade patterns. Vector texture pattern synthesis can obtain the rules of the elements' distribution from a given sample pattern and generate a large pattern which shares the same rules with the sample pattern. The element in those texture patterns can represent any independent visual information geometry or texture. The study of automatic synthesis method of these patterns has become an important topic, as these patterns not only plays a significant role in hand-painting simulation but also be used in virtual scene generation such as the generation of a forest. Furthermore, automatic synthesis of pattern means we use the computer, a machine, not a human being to analyze and understand the rules of the elements distribution, this involves in some artificial intelligence which also makes the study of vector texture patterns synthesis a more meaningful research topic. In this paper, a new sample based vector texture synthesis method is proposed. In our method, we put forward a Neighborhood Histogram for the comparison and matching of element's neighborhood. The neighborhood space of an element is divided into small grids by a bunch of radials and a series of concentric circles centered at this element. A histogram can be generated by counting the number of elements positioned in each grid. The histogram represents the neighborhood information of current element. It is used to search the matching element in sample pattern through histogram comparison, and hence synthesize the destination pattern with the matching element. The method includes three stages: sample acquisition, sample analysis and pattern synthesis. A sample pattern can be obtained by reading in a pattern file with specified format or being constructed interactively. We also need to obtain the elements information from the sample pattern, including position, types etc. The main task of sample analysis is acquiring the neighborhood information between elements in the sample pattern and building histogram for each element. We first estimate the radius of the element's neighborhood region according to the space arrangement among elements in sample pattern;then we will construct each element's neighborhood and build the histogram of the neighborhood. In the pattern synthesis process, first, we select an element in random, put it in the center of the synthetic area and copy its neighborhood elements to the corresponding positions in the synthetic area. The first element is named as the center element, and the center element and its neighborhood elements form an init pattern of our synthesis pattern. Next, we will extend the synthesis pattern step by step. In each extension, we choose the element nearest to the center element as an extending-element;then search the matching element in sample pattern whose histogram is the most similar to extending one's;copy the neighborhood elements of the matching element to the synthetic area. Repeating the process to extend the synthesis pattern, we can get the final result.We employ our method on regular, near-regular and random patterns respectively. We find that our method can accurately grasp the rules of sample pattern distribution and get good synthesis results. Most of the previous methods use Delaunay triangulation to get the relationship of elements in sample pattern, which means the relatively position of two close elements are defined in an accurate and fixed form. In synthesis process, as there is no element in sample pattern which shares the exact same neighborhood information with the one in the synthesis pattern, this usually causes difference accumulated and affects the final synthesis pattern. Some realization will slow down the speed of the program. Our method divides the neighborhood of an element into small grids and counts the number of elements positioned in each grid which used as a parameter in synthesis process. We use a certain range in relative orientation not an accurate relative orientation to calculate the neighborhood information of the element, making it some fuzzy in boundary, which can reduce the difference accumulation in synthesis process. Our method doesn't need complex elements relationship analysis and frequently comparison of elements positions, achieving higher performance. Our method can guarantee the local similarity between sample pattern and synthesis pattern, but as we didn't take the overall elements' density into consideration, this may lead to imbalance in element proportion. How to add this parameter into the synthesis process to achieve similarity between sample pattern and the synthesis pattern on the whole level is our first priority at present. For further extension, we could take vector field into consideration and generate effects with controllable gradual changes. In addition, our method doesn't involve in elements' parameters such as orientation, color, size and so on. Taking these factors into consideration and achieving good results with more expressive force is another extension we are trying to accomplish.关键词：vector texture pattern;pattern synthesis;non-photorealistic rendering(NPR);histogram matching4344|268|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