最新刊期
卷 19 ， 期 4 ， 2014
- 摘要：Due to the impact of the equipment and the environment, poor quality images could be generated inevitably in the process of image acquisition, which makes the later processing more difficult.In this paper,an edge preserving denoising method for color images through a controlling gradient vector is proposed.In order to make better use of the color image information,information loss is analyzed by transforming color images to gray image in the first step.Furthermore, the calculation method of a gradient vector in RGB space is proposed.Second,by analyzing regional structure differences of the noise gradient vector in a neighborhood and the edge gradient vector in a neighborhood,the inadequacy with diffusion equation is pointed out,and a method for controlling the diffusion matrix is proposed.Finally,PM equation in RGB space is proposed through the edge function derived by vector images,normal diffusion is removed by decomposition of model and improved edge stopping function with the controlling the diffusion matrix can make better control.The results show the proposed method has higher signal to noise ratio,and the effectiveness of it is verified.This method can denoise image,and it could keep the image contrast and edge information at the same time,so it has some practicability.关键词：image denoising;gradient vector;diffusivity equation;PM model;directional diffusion3298|440|0更新时间：2024-05-07
- 摘要：Low-rank matrix recovery is a hot topic in signal processing,artificial intelligence and optimization.Convex optimization based on low-rank matrix recovery problems coming from the compressive sensing technology,which is very popular these years for image processing,computer vision,text analysis,recommendation system,etc.Low-rank matrix recovery is achieved by minimizing the nuclear norm matrix to obtain a low rank solution,however,an unstable solution can be obtained due to the requirements for the low correlation of a low rank matrix.A low rank image denoising algorithm is proposed based on variable splitting method.The method introduces a Frobenius norm of low rank matrices as a new regular item and it is also combined with the original low rank nuclear norm to optimize the image denoising.In order to solve the improved denoising model,an augmented Lagrange multiplier method based on variable splitting is used by using convex relaxation of sparse recovery methods.Finally,to verify the stability and denoising capability of the presented approach, images with different noise types and simulation parameters are generated and processed using the presented method and the results are compared with the existing low rank matrix algorithm.Performance analysis of recovery time,signal-to-noise ratio,and error rate are evaluated at the same time.The proposed method can yield superior performance compared to the traditional low rank model in terms of the test results.The experiments indicate that the improved models, while keeping the original low-rank sparse recovery, have good denoising performance and exce-llent stability on the strong correlation matrix and we can get a higher signal-to-noise ratio.关键词：convex optimization;low-rank recovery;augmented Lagrange multiplier method;variable splitting;Frobenius norm4322|339|0更新时间：2024-05-07
- 摘要：Removing photometric discrepancy on seam-lines of 3D model textures, and achieve natural and continuous color transition across the texture seam-lines.We propose a new photometric processing method of face-wise Poisson blending,which is based on the standard Poisson image editing approach combined with color transfer techniques.First,we regard the texture triangle on images as the processing unit,and the color transfer parameters as the unknowns; then the Poisson equations set is constructed according to the edge adjacent topology on the triangular mesh; after that,the color transfer parameters are resolved using global least square; at last,color correction is locally performed with the global solution.Our method assures the smooth and continuous color transition besides the seams and simultaneously remoral of the global color difference between images more effectively.Compared with standard pixel-wise Poisson blending method,our procedure has the advantages in computing and implementation,which is more appropriate for seamless texture processing on complicated models.Even more,our approach has a better effect in global color difference correction and transition.Theoretical analysis and experiments prove the superiority of our method.关键词：texture mapping;seamless texture;photometric processing;Poisson blending;color transfer5354|445|0更新时间：2024-05-07
- 摘要：Texture is an important feature for describing different kinds of materials.Texture feature extraction is a hot topic in pattern recognition and computer vision research fields.LDP descriptor is a discriminative texture feature which is more robust to random noise and non-monotonic illumination variation than LBP.However,the LDP descriptor is much slow for calculating and sorting edge responses of 8 directions.To solve this problem, we improve the LDP coding method.We propose two improved methods.The first one adapts the same edge templates as LDP but uses a different coding scheme without sorting,which we called FLDP(fast local directional pattern).The second method uses less edge templates to construct short descriptor in order to reduce time and storage consumption of the feature,which we called MLDP(mini local directional pattern). We present experimental results on the Brodatz full set and Brodatz subset.Both show that the FLDP descriptor is 19 times faster than the LDP and the MLDP descriptors are 34 times faster with even better performance.Two methods are presented in this paper, FLDP and MLDP,to improve the LDP.Experiments show that these two improved descriptors cost much less time with even better performance than the LDP.关键词：texture classification;local directional pattern;texture feature;local binary pattern3133|321|0更新时间：2024-05-07
- 摘要：Reversible digital watermarking technology has been widely applied in many fields.Concerning most of the existing reversible watermarking algorithms need a lot of additional information; this paper presents a high capacity reversible watermarking algorithm, which does not need any additional information.This algorithm depends on the image interpolation space.First, we design a similar bilinear interpolation image amplification method,and enlarge the cover image by four times its original size. 1/4 pixels of the magnification image retain the original image information,and the other pixels can be embed the watermarking.During the interpolation process.We use a random function to select watermark-embedding position first and then determine the rounding mode and achieve watermark embedding by the watermarking information and the theoretical value.This method ensures a uniform distribution of the watermarking and it also minimizes the effect of the interpolation, while improving the security.Compared with the classical interpolation algorithms, the last image is better,and embedding rate can reach around 0.75 bit/pixel.Theoretical analysis and experimental results show that the algorithm is reversible and effective and secure.关键词：reversible watermarking;interpolate;bilinear interpolation;image magnification;information hiding3168|260|0更新时间：2024-05-07
- 摘要：Because of the complicated optical environment underwater and the moving target characteristics,it is difficult to extract target features and predict target sizes precisely in underwater videos.Thus, tracking window offsets become bigger and cannot envelop the target area accurately during the target tracking process.A novel approach of visual depth based target feature calculation and target tracking is therefore presented.First,visual depth information is calculated by dark channel prior,thus the target's spatial position feature is extracted.Second,dehazing and color restoration of the underwater image is applied based on the depth information and the target's feature will be enhanced.At last, an underwater target is tracked under the Bayesian filter framework.Meanwhile,the target window size is adaptively adjusted based on the target's spatial position feature.Experimental results show that the proposed algorithm can calculate target features and optimize tracking windows based on the visual depth.Thus, objects can be tracked adaptively.This paper presents a new underwater targets tracking method based on the calculation of visual depth information.Experimental results validate its robustness in underwater target adaptive tracking. Furthermore, it can be used in various nonlinear non-Gaussian underwater target-tracking frameworks.关键词：underwater target tracking;visual depth information;feature calculation;Bayesian filter3089|281|0更新时间：2024-05-07
- 摘要：Visual saliency refers to the physical,bottom-up distinctness of image details.It is a relative property that depends on the degree to which a pixel or region is visually distinct from its background.The region-based contrast method(RC) achieves good results on public dataset. However,the over-segmentation of the method is considered as a preview contrast computation which contributes to the high precision.In this study,a novel bottom-up salient object detection method based on color names and spatial information is proposed,in which regional contrast and spatial compactness are considered as two factors for saliency evaluation.In addition,we express the prior knowledge of traffic signs with top-down saliency maps,combined with the bottom-up saliency maps,to detect traffic signs from real world images.First,we learn color names offline by a probabilistic latent semantic analysis model to find the mapping between color names and pixel values.The color names can be used for image segmentation and region description.Second,pixels are assigned into special color names according to their values to form different color clusters in the image.The saliency measure for every color cluster is evaluated by its intra spatial variance.The less the color cluster spreads the more salient it is.Third,every color cluster is divided into some local regions which are represented by color name descriptors.The regional contrast is evaluated by computing color distance between different regions in entire image.Last,the final saliency map is constructed by incorporating the color cluster's spatial compactness measure and the corresponding regional contrast.Note that in most cases,it is not the bottom-up saliency,but the most "interesting" object in an image that attracts attention.In this study,road signs are divided into three categories; every category has special color information.For each category, a class-specific distribution is constructed by the bag-of-words(BoW)model with training images to form the top-down saliency maps.Then the traffic signs are detected from the saliency maps, which are generated by combining the bottom-up saliency maps with top-down saliency maps.When evaluated using one of the largest publicly available datasets,our method outperforms several existing salient object detection methods with an achieved accuracy of 92%.The ROC curve generated by our method is better than the curves produced by other methods with the area under curve(AUC)of 0.9453.In addition,when tested on the traffic sign dataset constructed by ourselves,our method achieves a detection rate of 90.7%.In the paper,we propose a novel saliency detection method,in which an image is treated as a composition of 11 basic color names.Every pixel belongs to different color names in a probabilistic manner.Based on this idea,the image is divided into some color clusters,followed by segmenting every cluster into local regions.For saliency measure evaluation,both spatial compactness of the color cluster and the region contrast are considered.Our approach achieves the best results compared with some state-of-the-art methods on the public dataset.Furthermore,it obtains good performance when considering task-dependent application for traffic sign detection.关键词：color names;visual saliency;region contrast;spatial compactness;road sign detection3844|329|0更新时间：2024-05-07
- 摘要：Traditional generalized Hough Transform in translation,rotation,scaling,partial occlusion and other circumstances,can locate any target, However, the slow positioning speed, the large storage space, the discrete accumulator space, and other problems hinder the usability of this method.Therefore, an improved generalized Hough transform based on a global adaptive artificial fish swarm algorithm is proposed to locate targets more quickly.According to the polar coordinates of the target shape information a reduced R-Table is established, which removes the gradient information to reduce the computational complexity and improve the robustness of the target model.Then we use a reduced R-table to calculate the value of the candidate target model as an artificial fish fitness value.And the artificial fish uses adaptive vision and step as well as constantly interaction and coordinate behavior to search the optimal target model parameters in the continuous multi-dimensional accumulator space heuristically which demarcates the exact location of the target.Experimental results show that,the algorithm requires only a constant level of storage space cost.The speed is improved more than 90% compared with the generalized Hough Transform algorithm.The algorithm not only greatly reduces the cost of space and time,but also improves the positioning accuracy of the target.We propose a new search strategy in the accumulator space,which can be more quickly and accurately to locate the target,especially the complex target in complicated backgrounds.关键词：generalized Hough transform;reduced R-Table;global adaptive artificial fish swarm algorithm;target location2683|234|0更新时间：2024-05-07
- 摘要：Global descriptors have the future of low dimension and simple similarity measurement.An improved global invariant is proposed based on the medial axis skeletons and geodesic distance for bending deformation.First, we use the geodesic distance to correct the image centroid, and then we extract the skeleton instead of the global image domain as operational point set to reduce the computational complexity. Finally, we propose single dimension invariants.The experimental results show that compared with other traditional feature representation the invariant is bending insensitive and meets the test of TRS(translation,rotation,scaling) invariance.It can effectively represent and identify bending non-rigid shapes.In this paper, we propose a novel global invariant of single dimension.Its simple structure makes it possible to use simple calculations to achieve real-time performance.This descriptor provides a new method to robustly identify objects with bending deformation.关键词：geodesic distance;articulated shapes;skeleton;invariant3301|230|0更新时间：2024-05-07
- 摘要：Because the digitalization of real objects often entails the texture mapping of the 3D geometry model, images-to-geometry registration is indispensable.The geometry of the target in close-range photogrammetry is complex, and the shooting angle of the image changes largely, which makes images-to-geometry registration difficult to process. Traditionally, a single image is registered with a geometric model by manually selecting the corresponding points. Thus for tens of images this processing can be very time consuming. For efficiency, the position and orientation of the multi-view images can be unified in one coordinate system at first. However, this process is also difficult because of the complex and uncertain background in close-range photogrammetry. To this end, we put a plane calibration board under the target while taking images, and use the camera calibration to get both, the internal and external orientation parameters; thus, the multi-view imagesare placed in one coordinate system. The unified coordinate system means when the registration parameters of one image are known, the registration parametersof all other images can be calculated through the relative positions and orientations among them. Then, at least three groups of corresponding points are manually selected, and the registration parameters are calculated by using the absolute orientation. The achieved registration parameters are not precise enough but can be used as an initial value. Finally, by making the normalized mutual information between multi-view images and the diffuse rendered maps of the geometric modelas a similarity measure, we use Powell's nonlinear optimization method to get the precise registration parameters.In the final process, we optimize all images together, while the traditional methods optimize each single image separately; the difference is we have only 7 parameters for all images, but the traditional methods have 7 parameters for each image. The experimental results show that using the calibration board we obtain robust internal and external orientation parametersof multi-view images, and after the absolute orientation process, we find obvious gaps between the images and the rendered maps. However, after a normalized mutual information based optimization process, the gaps are significantly improved.The typical total registration work takes about 10 minutes, in which there are 4 minutes for manual work and 6 minutes for computation; by contrast, the fully manual texture-mapping work by using the commercial software-Cinema4D takes nearly 4 hours. Registration between multi-view images and geometric model compared to registration between single image and geometric model, the workload of manually selecting corresponding points is greatly reduced.The errors of manually selected points can affect the accuracy of absolute orientation, but after using normalized mutual information based nonlinear optimization, we can still get precise registration parameters. The normalized mutual information based nonlinear optimization can also be used for single image-to-geometry registration, but for multi-view images the method is more robust. The optimization method has one drawback: it requires the geometric model has a high quality. This is because the normalized mutual information has a close relationship to the surface's normal vector of the geometric model. This drawback can be avoided by using high accurate 3D scanners.关键词：3D reconstruction;texture;geometric model;registration;mutual information;nonlinear optimization3515|261|0更新时间：2024-05-07
- 摘要：Since using limited spatial information from only one image patch to structure conventional binary descriptor such as BRIEF will result in low discriminative ability and is invariant to rotation or viewpoint changes. We work on this problem through further excavating feature information of the region of interest and by proposing a binary descriptor encoding, which does not only include intensity-comparison information but also information about relative relationship of intensities. This can be done by taking BRIEF-like tests on image patches gotten in two ways. First, several sub-patches can be obtained by decomposing one original image patch by sorting all pixels on it according to their intensity order, then random tests are used on each sub-patch, and all test results are concatenated to form a distinct binary string of sub-patches; second, multiple original image patches are produced by dividing interest regions into several patches of different sizes around the keypoint, or they can be retrieved from the scale space of image patches, and the discriminative power of the descriptor could be further raised by taking tests on these multiple patches. Results based on experiments of performance evaluation have shown that the proposed binary descriptor outperforms other binary descriptors or even floating point descriptors under various image transformations. Excavating feature information in local image region can improve the robustness of binary descriptors.关键词：keypoint;image patch;local feature;binary descriptor;robust3389|284|0更新时间：2024-05-07
- 摘要：Volume rendering is one important method for 3D data visualization. Volume data for volume rendering often includes a lot of empty voxels, resulting in decreasing the rendering efficiency for ray-casting algorithms, because meaningless resampling positions may be computed. This paper presents a GPU-based volume rendering method for full-empty sub-data blocks to efficiently improve the rendering speed. The sub-data blocks with a lot of empty space are effectively managed with an octree structure. By rendering surfaces of the non-full-empty sub-data blocks of leaf nodes in the octree,the start and end resampling positions of the ray-casting algorithm are closer to the visible part in the data volume. Meanwhile,the appropriate leaping steps are calculated to quickly skip the full-empty sub-data blocks according to the lookup result of the full-empty subdata blocks confirming texture. As a result, the meaningless resampling points are greatly reduced and the new algorithm, which is based on the selection of the depth of octree and effectively ski-pping full-empty subdata blocks, shows better performance than the original GPU ray-casting algorithm with bounding box surface rendering when the data volume contains a lot of empty voxel. The new improved algorithm is successfully applied in seismic data visualization, with an opacity function designed to highlight the layer surfaces in the data volume, which shows favorable practical effect.关键词：volume rendering;octree;full-empty subdata blocks;seismic data3055|230|0更新时间：2024-05-07
- 摘要：With the rapid development of telemedicine and online diagnosis,more and more precious medical images have been transmitted through the open network,and the security problem of medical images has been taken much attention by many researchers.The integrity authentication of transmitted medical image is a key step in information protection.Watermarking technology is generally regarded as an effective method for integrity authentication.However,like other quality-sensitive images such as medical imagery,any slight distortions are not acceptable,because the distortion may cause wrong diagnosis.Therefore,using lossless watermarking method for medical image's tamper detection and recovery is a very important research field.Aiming at the existent algorithm's defects in region division and block's features chosen,a new lossless watermarking method based on quad-tree decomposition and linear weight interpolation is presented in this paper.First,the quad-tree decomposition is used to divide the original medical image into flexible size blocks with high homogeneity.Because this division method can reduce the length of feature value,much more information can be embedded.Then a new linear weight interpolation method is used to compute each block's features as watermark information in our method.Because in existent methods, the mean used of the block's pixels are used as recovery feature for each block,the tampered regions occurring in the low frequency region of an image(usually,the image's texture complex area),it is not the best choice for restoring all tampered pixels with mean values, because it will reduce the quality of the recovered image.Finally,in order to reduce the embedding and extraction algorithm's complexity and enhance watermarking method's security,these features are embedded by using simple invertible integer transformation based on chaos.The introduction of chaos system improves the presented method's security.In order to test the presented method's validity,we test some performance properties including reversibility,embedding capacity,watermarked and recovered image's quality,and the detection accuracy rate.The experimental results show that the original image can be recovered with any distortion if the watermarked image was not been tampered,otherwise the tampered region can be localized and recovered with high similarity in the extracting procedure.Compared with the existent method,the embedding capacity is obviously prior to the similar method and the recovered image' quality using our method is about 20 dB higher.In addition,when the watermarked image is tampered to a large degree,the right detection rate and wrong detection rate is obviously better than the existent method.Since all the algorithms are open to the public,we introduce the logistic chaotic mapping to select the reference pixel's position in watermark embedding procedure.Due to the initial sensitivity and randomness,an adversary cannot obtain the correct watermark embedding positions easily if the initial parameters of logistic mapping are secret.Experimental results show that,compared with previous similar schemes,the presented method not only achieves high embedding capacity and good visual quality of restored image,but also has more accuracy for tampering detection.The presented method can be applied to realize medical image's integrity authentication and tampering detection.关键词：lossless watermarking;medical image;quad-tree decomposition;linear interpolation;tamper detection3017|352|0更新时间：2024-05-07
- 摘要：Many algorithms for aircraft recognition have been develop during the recent years. Traditional aircraft recognition algorithms are generally extracting the invariant features to train the learning machine after the targets segmentation, such as SVM, neural networks, and others. After training progress, the machine can recognize aircraft targets automatically. If there is less interference, traditional algorithm can work well. However, in actual circumstances, there are a lot of disturbances in the remote sensing images, such as noise, shadow, tanks, trees, etc. At this time, traditional algorithm cannot work well. However, for complex remote aircraft images, there is not a suitable segmentation algorithm for various aircraft models, so traditional recognition algorithm is not universal. Aiming at the deficiencies of the existing recognition algorithms, we propose an aircraft recognition method for remote sensing images based on the distribution of the feature points, color invariant moments, and Zernike invariant moments. Above all, preprocessing for remote sensing images and module images should be done. Preprocessing concludes a graying and noise-removing process. The graying process is making colorful images become gray images, which can draw Harris-Laplace corners and Zernike moments in the following process. After that, the next process is the noise-removal. Noise-removal is a very important procedure. If there is much noise in the gray images, the moments extracted from the images will not be stable. A wavelet is a wave shape, which has a small area, finite length, and zero mean value. By changing the coefficients, wavelet transform can refine the input images with multiple-scales, and it can analyze signals in the time domain and the frequency domain at the same time, so it is called a mathematical microscope. When the preprocessing is done, the remote sensing images and module images are transformed by a wavelet. In the low-resolution images, use ring projection for rough matching. In the high-resolution images, we extract multi-scale corners using Harris-Laplace, and triangulate use Delaunay-triangulation. At the same time, abstract color invariant moments and Zernike moments. Afterwards, take the Euclidean distance as the similarity measure for the three different features, and use these features to match the sample images with the weighted coefficients. At last, we select the sample image, which has minimum Euclidean distance to the finally recognized target. After experiment, the results show that aircraft detection accuracy in this paper is 2.2% higher than recent algorithms, and aircraft recognition accuracy is 1.4%10.4% higher than recent algorithms. This algorithm can accurately identify ten aircraft targets in remote sensing images, and has good robustness to background, noise, viewpoint changes, and other interference. In this paper, we propose a new aircraft target recognition algorithm based on the spatial distribution of feature points, color invariant moments, and Zernike invariant moments. There are three different features in this algorithm, containing feature points and invariant moments, which include diverse information of the images. Therefore, the new algorithm overcomes the shortcoming that the single feature cannot contain enough information. The experimental results show that the new aircraft target recognition algorithm in this paper has better anti-interference ability and recognition accuracy than other algorithms.关键词：aircraft target recognition;remote sensing images;distribution of the feature points;invariant moments4167|376|0更新时间：2024-05-07
- 摘要：With the rapid development of remote sensing satellites, the size and the resolution of satellite images is growing increasingly. The evaluation of remote sensing image quality requires precise information of control points extracted from unevaluated images and reference images. Therefore, we propose an adaptive Wallis enhancement method based on sparse recognition to increase the number of control points and to improve the matching precision for high-resolution images. First, feature vectors of images are constructed by computing the image radiation-parameters. Second, the classification of sub-region terrain in the image can be determined using sparse recognition. Finally, according to the specific type of the sub-region terrain, we enhance the regions by the Wallis filter adaptively based on corresponding filter parameters and extract control points which would lead to the automatic evaluation for geometric precision. The experiments show that the proposed method can get better results especially in the detail on ZY-3 images, hence can increase the number of and improve accuracy of control points. In this paper, we propose a new enhancement algorithm for high-resolution remote sensing images that we enhance textures of sub-region terrains using adaptive Wallis filter. Compared with traditional Wallis filter, our method eliminates the number of pixels with saturated gray to increase the number of and improve accuracy of control points.关键词：sparse recognition;radiation-parameters;adaptive Wallis enhancement;extract control points4346|287|0更新时间：2024-05-07
- 摘要：To improve the robustness of transform domain watermarking and solve the limits of the grid shape in multi-resolution wavelet transform, a transform domain-watermarking scheme of 3D models based on local feature points is presented in this paper. 3D models are partitioned through longitude and latitude angles to get the local feature points. Two-dimensional matrices are composed by the ratio of the length of the local feature points and the other points in this region. The discrete wavelet transform is carried out on the two dimensional matrix modifying the wavelet coefficients of each layer in the intermediate frequency and high frequency coefficients to embed the watermarking. Then, the airspace signal is obtained by the inverse discrete wavelet transform and the length of the local feature points is modified to obtain a 3D watermarked mesh model. The proposed watermarking scheme is robust against attacks such as translation, rotation, uniform scaling, vertex re-ordering, noise, simplification, cropping, subdivision, quantization, combination attacks and so on. The experimental results show that this watermarking scheme has good robustness and imperceptibility. Besides, it has no strict requirement for the shape of the mesh model.关键词：3D watermarking;3D mesh model;local feature points;transform domain;discrete wavelet transform(DWT)4339|320|0更新时间：2024-05-07
- 摘要：With the explosive growth of internet information, although search engines can provide convenient information retrieval services, people still have to spend a lot of time and energy to find the information they are interested in from millions of search results. Therefore, it is significant to mine user's preference from the search engine's interactive information and to provide personalized search service. Our method is based on a systematic analysis of current researches on user preference mining. We propose a novel preference model named HWUG, which fully considers the relationship between the user preference information. It can extract preference label and action from the user's explicit and implicit feedback information. A time attenuation factor is brought into the historic preference information in order to provide the real-time computation of preference information. The proposed methods have been used in the soccer video search engine(www.findballl.net)and achieved good results in personalized video retrieval and recommendation. The experimental results show that the combination of domain-specific knowledge, based on a hierarchical graph model undirected weighted preferece mining algorithm can more accurately reflect the user's soccer preferences in real-time.4977|236|0更新时间：2024-05-07
- 摘要：With the explosive growth of multimedia data on the internet, the efficient organization and retrieval of large-scale image and video data has become an urgent problem, which expects more efficient low-level feature with low computation. This brings a huge challenge to the conventional visual feature. It is urgent to make descriptor more compact and faster and meanwhile remain robust to many different kinds of image transformation. To this end, we first introduce several schemes of binary features, and then propose a novel fast descriptor for local image patches. A string of binary bits is used, which are derived from the intensity difference quantization between pixel pairs that are sampled according to a fixed random sample pattern. Different with the other binary descriptor approaches, our method first extracts the pixel pairs randomly, and then calculates the intensity differences from these point pairs. We quantize these intensity differences into binary vectors to form the local binary descriptor. Our experiments show that our method is very fast to extract and it shows better more robustness than the other binary feature schemes. The binary descriptor proposed inthis paper is computed very fast and it outperform other binary features on the public datasets we used. It proved that quantization-based method can obtain more robustness than compare-based methods.关键词：multimedia technology;local feature;image retrieval;binary features4448|284|0更新时间：2024-05-07
- 摘要：Mesh denoising is a typical problem in computer graphics. The key challenge we face in this field is to denoise the mesh and maintain the structure of the mesh at the same time. And it is becoming the hottest topic in this area. We propose a global mesh denoising method using -sparsity. This method is motivated by the fundamental theory of sparse representation in the field of signal processing. The global optimization of an energy function is used to remove noises from the mesh while the features are preserved. There are two steps in our method. The first step is the filtering of the face normals. We formulate a global optimization model to optimize the face normals of the noised mesh. Then we use the -norm to ensure the sparsity of the solution, which preserves the structures of mesh features. The second step is the reconstruction of the denoised mesh. Given the new filtered face normals, we create a vertex reconstruction model under the least-square sense according to the definition of the face normal. The denoised mesh is updated by the solution of the reconstruction model. Furthermore, our model solves the denoising problem globally, which avoids problems of existing methods, such as the convergence problem. A large number of experiments demonstrate that our method is able to remove noises, at the same time, preserving the features of the mesh, especially for the CAD models.关键词：computer graphics;mesh denoising;global filtering;-sparsity4903|270|0更新时间：2024-05-07
- 摘要：The MVC standard, as an extension of H.264 standard, provides a standardized codec reference for 3D videos. Real time encoding and keeping compatible with H.264 are key problems for deploying MVC in 3D applications. In this paper, a transformation mechanism is proposed based on topological sort, which maps the 2D prediction structure of MVC to 1D reference sequence of H.264, and then uses the basic coding technologies of H.264 to get MVC bit stream with syntax reconstruction. The experiment results show the proposed method could achieve real time performance for MVC-H.264 transcoding and are compatible with existing H.264 codec. The proposed method gives transformation mechanism of bit streams between MVC and H.264, which need no special designed SOC.关键词：H.264;multiview video coding;transcoding;parallel encoding4912|245|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