最新刊期
卷 18 ， 期 11 ， 2013
- 摘要：With the rapid development of computer network, social media, digital television and communications technology, generating, processing and access to multimedia data has become more convenient. Multimedia applications are increasingly widespread, and the amounts of multimedia data are showing the explosive growth. In the era of big data, multimedia data has become a major data objects. However, due to its unstructured nature of multimedia data, the processing and retrieval of multimedia data is relatively difficult. How to effectively store, organization and management of these data, how to access and retrieve data in accordance with the multimedia content and features, has become an urgent need. This report is organized around video event detection and annotation, high-dimensional index structure and cross-media search three research directions, and elaborates their development status in 2012 and future development trend outlook.关键词：multimedia;event detection;index;cross-media;retrieva2865|328|0更新时间：2024-05-07
- 摘要：In recent years, image processing based on the partial differential equation has stronger local adaptive characteristics and a high degree of flexibility. This method becomes a new research hotspot after image processing based on wavelet. Based on the different characteristic and superiority represented by the thermal diffusion model and the TV(total variation)model in the image smooth area and edge or texture area,we propose the adaptive hybrid-image magnification model based on image local characteristics. Initialized by the amplification image after the double linear interpolation, this model adaptively adjusts the diffusion model according to the pixel gradient characteristic in the smooth area and edge or texture area. The proposed model makes the isotropic diffusion in the smooth area, and the direction diffusion along the vertical gradient direction in the edge or texture region. This model inhibits the massive effect and the false edge phenomenon brought by the image interpolation amplification. The proposed model is proved theoretically, and at the same time, a lot of simulation results verify the effectiveness of the proposed model.关键词：image magnification;bilinear interpolation;thermal diffusion model;total variation(TV) model;adaptive hybrid model2636|227|0更新时间：2024-05-07
- 摘要：Salt & pepper noise is one of the key factors causing image contamination. Its removal is a research hotspot in image denoising. Directional weighted median filters face some issues when seeking restored gray levels of noise pixels, such as non-exclusion of noise neighbors' disturbance, inaccurate estimation for direction and incomplete depiction of local gray level characteristic. To alleviate the issues, a directional weighted mean filter is proposed. The proposed algorithm first detects noise pixels by using directional gray level differences and extreme judgment, then adaptively takes the weighted gray level mean of recursive or non-recursive filtering window as restored gray levels of noise pixels according to noise density estimation of local window. Simulation results show that, as compared with two existing directional weighted median filters, PSNR values of the proposed algorithm is usually increased by 23 dB and 56 dB, with a more obvious increment for high noise density. In addition, running speed is increased nearly 10 and 30 times.关键词：image processing;salt & pepper noise;mean filter;median filter3373|247|0更新时间：2024-05-07
- 摘要：Since existing algorithms cannot solve the threshold segmentation problem of mixed noise images a 3D minimum error threshold algorithm is proposed in this paper. Using gray distribution information of pixels and relevant information of neighboring pixels, it combines image gray, mean and median values to construct a 3D observation space, and then defines a 3D optimal threshold discriminant based on the relative entropy. Furthermore, in order to improve its processing speed, the fast calculation method based on decomposition is proposed. It calculates three 1D optimal thresholds, instead of one 3D optimal threshold. Its time complexity is reduced to , and space cost is reduced to . Experimental results show that the proposed algorithm outperforms those 2D threshold methods for different types of noised image and non-uniform illuminating images. Especially for mixed noise image, its advantage is more obvious.关键词：image processing;threshold segmentation;minimum error threshold method;three-dimensional gray-level histogram;relative entropy3122|256|0更新时间：2024-05-07
- 摘要：In this paper, we discuss a way of image inpainting with large defect of structural information by structural constraints and sample sparse representations. The image edge information is repaired by a polynomial curve fitting to constrain the structural information. First, a narrow-band model of sample sparse representation is used to repair structural information. Then, the texture information is completed by a translational block sparse representation method. Simulation results show that proposed method can achieve higher image quality, and can better repair the structure information and maintain the smoothness of structure integrally.关键词：sample sparse representation;structural constraints;polynomial curve fitting;narrow-band model;translational block2876|272|0更新时间：2024-05-07
- 摘要：A new fast fusion algorithm for multi-focus images based on wavelet transform and adaptive block is proposed in this paper. The proposed algorithm is implemented under the framework of wavelet transformation. For the low frequency coefficients, an adaptive block-based fusion technique is applied, where the optimal block size can be calculated by using differential evolution algorithm. Moreover, a pixel-level label map, which can accurately indicate the origin information of each pixel, is obtained by refining the above low-frequency fusion-result. On the other hand, the high frequency fusion task is completed by combining the local wavelet energy based rule with the information offered by the label map. Finally, the fused image is obtained by performing the inverse wavelet transform. Experimental results demonstrate that the performance of the proposed method is comparable with the state-of-the-art methods on both subjective visual perception and objective evaluation criteria. Furthermore,the proposed algorithm can achieve a good balance between improving the fusion quality and reducing the computational cost.关键词：multi-focus image fusion;wavelet transform;differential evolution algorithm;focus measure4175|343|0更新时间：2024-05-07
- 摘要：To solve the problem of low accuracy of multi-view scattered point cloud registration in objects surface 3D point cloud reconstruction process using TOF (time-of-flight) cameras, this paper presents an optimization registration method. The method obtains the absolute transformation matrix of any point-cloud to the reference coordinate system by building a target functions and combining adjacent point-cloud transformation matrix to minimize the objective function, avoid the accumulation of errors caused by registration of continuous point clouds. We use different objects for our experiments, and the results show that the method ensures the speed of point cloud registration. At the same time, it improves the accuracy of the multi-view point cloud registration, as well as enhances the effect of the objects point cloud model reconstruction. Furthemore, it is conducive to realize 3D surface mesh reconstruction.关键词：multi-view optimization registration;minimized objective function;global optimization;TOF camera3395|702|0更新时间：2024-05-07
- 摘要：Skin lesion detection plays dominant role in computer-aided diagnosis system of dermoscopic image. Be different from existing segmentation algorithms which are based on traditional features like texture or color, the proposed algorithm is based on independent pigment concentration distribution. First, a dermoscopic imaging model is built based on Lambert-Beer law, followed by obtaining and visualizing pigment concentration distribution after applying ICA method. Then give the definition of Pigment Concentration Ratio(PCR), which puts emphasize on gradient distribution of single pigment concentration. Final segmentation is created by applying global or local thresholding-based algorithm on PCR. Experiments show the robustness and generalization of proposed algorithm. Especially, expected segmentation result is achieved even if lesion regions of input image show very dim gradient.关键词：pigment concentration distribution;melanin;hemoglobin;skin lesion segmentation;dermoscopy2736|280|0更新时间：2024-05-07
- 摘要：GPUS can deal with massive data quickly and efficiently. Therefore, in recent years, it becomes a research hotspot in the field of images processing.The existing GPU rendering algorithms have a low utilization rate of resources and excessive bandwidth consumption in dealing with scene models containing a lot of the same or similar models. Therefore, a method of accelerated rendering based on CUDA is proposed on the basis of the original shader architecture in the GPU.In our method, building the corresponding model according to the existing GPU rendering model, and then finding out its deficiencies through model thus derivate constant memory concepts. Then analyzing the features of the constant memory and its effects on the rendering, thus deriving a method based on constant memory to accelerate rendering. The whole rendering process can be controlled by the rendering algorithm.The experimental results show that this method has a good effect to solve the above problems and accelerate rendering finally关键词：render;graphic processing unit(GPU);parallel computing;compute unified device architecture(CUDA)4165|266|0更新时间：2024-05-07
- 摘要：The max-flow model is used as a powerful tool to image segmentation problems. In recent years, a new model based on the continuous max-flow approach is presented and well adapted to segment images. However, the constraint of spatial flow in the model is simply set to a global constant, and it is not associated with architectural feature. Meanwhile, the initial terms of the source flow and sink flow need large amount of calculation, and cannot obtain satisfactory segmentation results. In order to overcome these shortages, we combine architectural features and statistical features of the image and a preprocessing algorithm, which include piecewise constant algorithm and Otsu-Histogram algorithm. Experimental results verify that the model is efficient.关键词：max-flow;continuous;image segmentation;piecewise constant3081|267|0更新时间：2024-05-07
- 摘要：Against the problem that compression algorithms for bag-of-features(BoF) ignore the spatial relationships of coded vectors, we propose a fusing algorithm of compression algorithms and spatial pyramid model in this paper. Meanwhile, we carried out a set of comparative experiments on several public image datasets. The experimental results show that compression algorithms are robust to visual word numbers and pooling methods of coded vectors. Otherwise, compression algorithms based on subspace methods have achieved best classification performances in the high-level feature space, and best accuracies and smallest time cost in multiple image datasets.关键词：bag-of-features;topic model;subspace method;dictionary learning;image classification2923|270|0更新时间：2024-05-07
- 摘要：Inspired by objectness measurement and visual saliency, we propose an object window center-surround depth distribution hypothesis for single-view image 2D-to-3D conversion. Based on this model, a depth estimation method fusing objectness and visual saliency is presented. First, visual saliency detection is performed and it is mapped to the depth. Second, some windows are sampled randomly for objectness measuring. Third, an energy function is used to model relations between depth and objectness, and then it is minimized by an iterative method to improve depth estimation results. Finally, 3D video is rendered based on depth. Experimental results show that our method improves depth discontinuity at objects boundary and continuity at other regions greatly.关键词：objectness measurement;visual saliency;2D-to-3D conversion;3D video3166|233|0更新时间：2024-05-07
- 摘要：In order to ensure the realism of visualizing complex forest scenes and maintain the constancy of generating dynamic forest scenes, this paper presents an approach to the adaptive visualization of multi-style composition for complex forest scenes. The approach determines the mixing ratio of tree models according to model contribution functions based on the range of visibility, which establishes the multi-style composition visualization model of the forest scenes. After estimating the calculation time of the tree growth model and the rendering time of the 3D models, as well as calculating the importance of trees in the forest scenes, the best scheme is obtained to generate the forest scenes. It keeps a relatively good stability of generating forest scenes, and the rendering strategy of forest scenes is dynamically adjusted according to the simulation results. To verify the effectiveness and practicality of this approach, we applied it to the dynamic forest simulation scene. The application results show that this adaptive visualization of multi-style composition can achieve great visual reality and improve the efficiency of rendering rates roaming, which makes forest scene roaming more stable and smooth.关键词：virtual forest;adaptive visualization;multi-style composition;estimation of simulation time;importance of object2327|318|0更新时间：2024-05-07
- 摘要：In order to reconstruct high-resolution images of sea surface temperature (SST) preserving the structural details using less amount of data, ridge and valley lines of SST,ridge lines of SST gradient and ridge lines of residual image were sampled. Combined with the land boundary and rectangle domain edges, the sampled points were set as boundary conditions.By the 9-point difference scheme of the Laplace equation, all nodes were calculated using finite-difference iterative method and the SST field was reconstructed. As a test, the SST image of level 3 MODIS product covering East China Sea and West Pacific in May 2012 was resampled and reconstructed. The result showed that the maximum absolute error between the retrieved image and the original one is less than 0.396 0 ℃. The reconstructed contours fit well with the original ones in the overlapped picture. It showed that the method for SST field reconstruction is feasible.关键词：structural detail;sea surface temperature(SST);sampling and reconstruction;finite difference scheme2586|231|0更新时间：2024-05-07
- 摘要：In this paper, we propose a color balancing method for correcting uneven illuminations of sub-images of a combined four-head camera. First, the radiometric response function is estimated via a process of camera calibration. Then, based on the overlapped areas of the four sub-images a color correction model is established. Furthermore, a robust estimation method is used to compute the differences of exposure amounts pertaining to the four sub-cameras. Hence, the process of color balancing for a stitched image is accomplished. Finally, a linear weight function is defined to smooth the stitching line. The experiments show that the proposed method is capable of eliminating or compensating the differences of exposure amounts, which guarantees the color unity of the stitched image without unacceptable boundaries.关键词：stitched image;color balancing;adjustment;radiometric response function;robust estimation2820|248|0更新时间：2024-05-07
- 摘要：Suffering from the interference of all kinds of background factors (clouds, shadows and pollution) and complicated changes of water itself in spectrum and shape, the process of water extraction can still not be completely automatic. Based on the goal of automatic interpretation of diverse water bodies under complicated backgrounds a new water extraction process is put forward by practical requirements. Adaptive segmentation or classification, local iteration, and post processing are the key steps of the proposed method mainly designed for the automatic extraction of large-scale water body information from Landsat TM/ETM images. The experiments are conducted on two typical research areas, Balkhash Lake and the Yangtze River region, each with 8 scenes of TM images. And we found that the proposed method can effectively overcome the interference factors, such as the cloud, shadow, water changes, which can achieve better extraction effect and preliminarily satisfy the actual application needs.关键词：large-scale water extraction;automatic interpretation;Balkhash lake;Yangtze River;adaptive iteration3095|272|0更新时间：2024-05-07
- 摘要：Realistic and immersive 3D tree modeling is an important and challenging research topic in computer graphics,computer vision, and virtual reality.Furthermore,it is highly valuable in various applications such as virtual tourism,virtual city,virtual agriculture,ecological landscape, and so on.Many researchers in this field used various approaches of 3D tree modeling.However,a review work is not available on this topic,which has obviously blocked the development of image based modeling research.Different methods are classified and analyzed based on the visual clues used in modeling from a computer vision perspective.In this paper,all advanced 3D tree modeling technologies are surveyed,which are categorized as image-based,rule-based,and sketch-based methods.After analyzing and comparing several main 3D modeling methods and newly key modeling technologies,Such as approximate image-based tree-modeling using particle flows,single image tree modeling,sketch-based tree modeling using Markov random field,texture-lobes for tree modeling,image-based lightweight tree modeling,we draw several conclusions.Rule-based tree modeling is based on the assumption that the branch shapes and phyllotaxis of the tree follow a predictable pattern,but it is difficulty to reconstruct the unpredictable part of branches such as stunted growth or suffered external environment influence.Therefore people should master large professional knowledge of botany to reconstruct realistic rule-based tree model.When it comes to sketch-based tree modeling,the users usually stretch the outline from the tree images to obtain the seeds,which guide the growth of trees or morphological characteristics.Tree modeling is relatively simple and intuitive and even can be realized by automatic or semi-automatic way.Image-based tree modeling is a method that 3D space geometry information of the tree is built from two or more than two images.It can be built intelligently and the model is realistic.But the model data is too large to meet 3D visualization display of the large-scale virtual scene on the Internet.In addition,a research idea about hybrid lightweight tree modeling for lightweight,low cost,high efficiency,which is proposed according to the current research emphasis and difficulties.At last,the future of lightweight 3D tree modeling applied in the interactive virtual scene is presented.关键词：3D tree modeling;image-based tree modeling;rule-based tree modeling;sketch-based tree modeling;light-weight modeling;tree skeleton5350|251|0更新时间：2024-05-07
- 摘要：The issue of copyright protection for 3D models is increasing prominently with the development of information technology.A non-blind watermarking algorithm for 3D mesh models based on mesh segmentation is presented.First,meaningful mesh segmentation on 3D mesh model is conducted using the mesh segmentation algorithm based on shape diameter function.Then,the robust center of gravity for each sub-block is calculated which is used as the center to transform the model to the spherical coordinate system from the Cartesian coordinate system.Finally,watermark sequence is embedded by modulating the distribution of vertex norms for each sub-block.In watermark detection,a non-blind method is used to detect the watermark sequence.Aiming at the problems of inconsistent mesh segmentation and the strong dependence on the patch boundary in current watermarking algorithms based on mesh patching,a mesh segmentation algorithm is introduced in and a block matching process with an original model is added to the register and resample process to ensure the consistency of the mesh segmentation.Furthermore,the distribution of vertex norms for each sub-block is chosen to be the watermark embedding primitive,which can weaken the dependence on the patch boundary.Experimental results show that the proposed algorithm can resist a variety of common attacks including translating,rotating,uniform scaling,noise adding,mesh subdivision,mesh simplification,mesh cropping and their combined attacks.关键词：non-blind watermarking;mesh segmentation;distribution of vertex norms;robust watermarking3930|196|0更新时间：2024-05-07
- 摘要：Color transfer between colorful images has important applications in the area of image processing and digital edutainment.In this paper,the essence of color transfer is investigated based on Reinhard's method,a new method for color transfer with multi-parameters is proposed by combining the scaling and mean values.Various choice of scaling values is presented and the mean value of the reference image is changed into a value as a linear combination of mean values of reference image and source images.By the proposed method,various different colorful images with different visual results can be obtained from given images,which provide more choices for users.Then the proposed method is extended to color analogy between multiple colorful images.Several examples are presented to show the effectiveness of the proposed methods.关键词：color transfer;statistics method;color analogy;color transfer with multi-parameters4148|279|0更新时间：2024-05-07
- 摘要：Pastel painting is a very expressive way of painting.This paper presents a non-photorealistic rendering method for pastel-simulation based on input images.The simulation is carried out in two stages,namely foundation-laying and refinement.In the foundation-laying stage,the input image is segmented by a mean-shift segmentation method to obtain regions and their colors.The vector field is calculated within the regions to determine the direction of strokes.Controlled by the above information,the foundation-laying stage is completed by simulating line-rendering and side-rendering techniques.In the refinement stage,prominent features are located in the reference image using the coherent line drawing method,and simulated with stipple-rendering in the pastel.Experiment results demonstrate that the two-stage process proposed in this paper transfers reference images quickly and efficiently into their pastel simulations.关键词：non-photorealistic rendering;pastel;image segmentation;vector field;coherent line drawing;stroke simulation4141|248|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