最新刊期
卷 20 ， 期 10 ， 2015
- 摘要：To solve the inadequate imperceptibility of reversible video watermarking and lacking of tamper localization, we present a novel reversible watermarking algorithm based on an intra-prediction mode. The proposed algorithm elaborately analyzes the sensitivity of the luminance prediction mode to different types of tampering of the I frame, extracts the prediction mode of each intra macro block, and generates a signature with the prediction mode. According to the features of a H.264/AVC codec, the minimum error coefficient in each 4×4 luminance residual block is selected with an error compensation method. Finally, feature information is embedded as watermarks in the selected coefficient by employing a difference expansion. Experimental results show that the original video can be recovered without any distortion if the watermarked video is not tampered; otherwise, the tampered region can be localized accurately, and the tamper localization accuracy is 4×4 blocks. Selected minimum error coefficients embedded with watermarks effectively reduce their impact on video quality. Compared with existing methods, the proposed algorithm improves the PSNRs of watermarked images by an average of 10%. Furthermore, the bit increase rate of the test sequence decreases by an average of 22%. Experimental results show that compared with previous reversible watermarking schemes for H.264/AVC, the proposed method not only achieves higher embedding capacity and better imperceptibility but also features greater accuracy in tampering detection. The proposed method can thus be applied in many fields, including the areas of medicine, military, and satellites.关键词：videowatermarking;H.264/AVC;reversible watermarking;difference expansion;intra-prediction mode3754|1112|2更新时间：2024-05-07
- 摘要：To address the drawback of the staircase effect of the total variational method,we proposed a weighted second-order total generalized variational model for wavelet inpainting. Unlike the total variational method, the proposed model contains a second-order derivative term and a first-order derivative term, which the model can automatically balance with two regularization parameters. To utilize the local structure of image information, we introduced an edge indicator function in the proposed model. The edge indicator function was 0 when the pixels belonged to the edge domain of the image and 1 when the pixels belonged to the smooth domain of the image. Thus, the proposed edge indicator function can preserve the edges and fine parts of restored images while improving noise removal in the smooth domain. To compute the new model effectively, we introduced a new variable and then used the alternative direction method to convert the original model into two submodels. For the first submodel, we used variational theory to solve the energy function and obtain the corresponding closed solution. For the second submodel, its non-convex characteristic made the derivation of solution considerably difficult. We sought to overcome such disadvantage by introducing the iteratively reweighted method. This method was employed to convert the original non-convex problem into several convex ones, that is, we improved the edge indicator function using the last restoration image and turned the non-convex submodel into a convex one. Subsequently, we introduced dual variables and transformed the second submodel into a minmax problem, which we then solved using a primal-dual algorithm. The comprehensive experimental results show that the new model obtains better results than recent total variation regularization wavelet inpainting methods.The average values of PSNR, MAE, and SSIM obtained with the total variation method are 21.884 4, 6.857 8, and 0.827 2, respectively; whereas, the values obtained with the proposed method are 22.313 8, 6.626 1, and 0.831 8, respectively. Total variation regularization is widely used in image processing. However, this technique produces the staircase effect. To overcome this drawback, we proposed a new weighted variational model for wavelet inpainting. This new model contains a total generalized regularization term that includes first-order derivative and second-order derivative terms. This feature allows the new model to automatically adjust the terms through two regularization parameters. To solve the new model, we first used the alternative direction method to transform the original problem into two subproblems. For the first subproblem, we obtained the necessary closed solution. For the second subproblem, we used the iterative reweighted method to reduce this subproblem into a convex problem, which we then solved using a primal-dual algorithm. For fair comparison, we used Daubechies 7-9 bi-orthogonal wavelets with symmetric extensions at the boundaries. The level of wavelet decomposition was 3, and the low frequency wavelet coefficients were lost randomly. We adopted a signal-to-noise ratio in decibels, mean absolute deviation error, and structural similarity to evaluate the quality of the restoration image. Experimental results show that compared with existing state-of-the-art algorithms, the new model is more effective in wavelet domain inpainting and is capable of obtaining better inpainting results for certain smooth images.2885|281|1更新时间：2024-05-07
- 摘要：Existing algorithms for detecting image copy-paste forgery can identify pairwise-similar regions in suspicious images but cannot accurately locate tampered regions. To address such drawback, we proposed a novel method for the automatic detection and localization of tampered regions on the basis of the offset estimation of double JPEG(Joint Photographic Experts Group)image compression. First, key points and their corresponding feature vectors were extracted using the SIFT(scale invariant features tromsform) algorithm, and a preliminary match of the feature vectors was implemented with the nearest neighbor algorithm. Second, the mismatch caused by inconsistent color information was eliminated by combining the preliminary matching points with the HSI(hue saturation intensity) color features to optimize the initial matching key points. Third, the matched key points were used to estimate the affine transform parameters and to eliminate the mismatching points with the RANSAC(random sample consensus) algorithm. Fourth, all the pixels within the copy-move regions were obtained by building a region correlation map. Finally, the duplicated region and the forgery region were distinguished from the copy-move regions according to the estimated offset of the double JPEG image compression. Compared with the classical SIFT and SURF(speeded up robust features) detection methods, our method can achieve higher true positive rates and effectively reduce false positive rates. When the second JPEG compression quality factor is greater than the first, the rate of forgery region detection can reach more than 96%. The proposed approach can effectively locate the forgery regions of JPEG images counterfeited via copy-paste tampering. It also demonstrates excellent robustness for copied regions distorted by geometric transformations and common post-processing operations.关键词：JPEG images;double compression;offset estimation;region duplication;forgery detection2898|564|5更新时间：2024-05-07
- 摘要：To overcome the deficiency of local gradient coding, which only extracts texture feature in neighborhoods of a fixed size, we propose a novel multi-scale local gradient coding fusion method based on asymmetric regions for feature extraction of facial expressions. A normalized face image is preprocessed by using a Gaussian filter to reduce the impact of noise. Then, the preprocessed expression image is divided into several blocks. For each pixel of each sub-block image, multiple and differently sized operators of local gradient coding based on asymmetric regions are used to obtain two binary sequences. These binary sequences are fused into a new binary sequence according to the logical XOR. The new binary sequence is then encoded, each sub-block histogram distribution is statistically analyzed, and all the sub-block histograms are cascaded into the texture features of a facial expression. Finally, the process of expression classification is completed with the SVM method. Experiments using the proposed method are performed using the JAFFE database and CK database. The average recognition rate for JAFFE is 95.24%, whereas that for CK is 96.83%. The proposed method is compared with LBP, CBP, LGC, and AR-LBP. Experimental results demonstrate that the proposed approach for the JAFFE database achieves recognition rates that are 5.6%, 4.85%, 3.71%, and 2.40% higher than those achieved with CBP, LBP, LGC, and AR-LBP, respectively. As for the CK database, the proposed approach achieves recognition rates that are 3.66%, 2.50%, 2.17%, and 1.66% higher than those achieved with CBP, LBP, LGC, and AR-LBP, respectively. Cross validation results show that the proposed method for facial expression recognition has excellent accuracy. Through the fusion of the intensities between neighborhoods of different gradients and scales, the multi-scale local gradient coding fusion method based on asymmetric regions combined with block histograms can perform well in local and global feature description. Experiment results show that the proposed method is better than typical feature extraction algorithms and is suitable for static facial expression recognition.关键词：expression recognition;asymmetric region local gradient coding;feature extraction;multi-scale fusion;SVM3215|448|3更新时间：2024-05-07
- 摘要：Edges are important image features that serve as basis of follow-up measurement and shape registration. To achieve good edge information, an edge detection method of block distance combined with summed area table is proposed in this work. The major innovation points of this study are as follows. 1) The edge of an image is detected by local block distance. 2) The sum of block pixels is accelerated by an integral diagram, and the method for completing a Gauss template block with an integral map is modified to improve the execution efficiency. The pixel difference of each block is computed. The pixel difference values are then accumulated to detect the edge of an image. The principle of this method is as follows: 1) the difference accumulation for gray regions tends to be zero, and 2) the difference accumulation for edge regions is different. The neighborhood features are then considered in a small range by comparing the differences among all the pixels in the adjacent region to determine the gradient of the central position. The structure of the Gauss template is analyzed, and the algorithm execution efficiency is improved by introducing an integral diagram. The size of the Gauss template is determined according to the rectangular area by using the integral diagram. The elements of the integral diagram, which are all ones, are constructed on the basis of the size of the Gauss template, whereas those, which are all zeros, are constructed on the basis of the size of the rectangular area. Then, the matrix of ones is used to achieve ergodicity for the matrix of zeros. Subsequently, the traversal times of each unit in the rectangular area after the completion of the traversal are obtained. A matrix of the traversal times could thus be formed and is then decomposed into the matrix of multiple ones until further decomposition is no longer possible. According to the matrix block of ones obtained after the decomposition, the pixels of the central points could be accumulated by the integral diagram. Finally, the overall distance value is remapped as the result of the edge. Unlike manual results, the results obtained with the proposed method indicate an overlap rate that is higher than 97%, which is greater than the overlap rate of the Canny edge detection algorithm (less than 80%), as well as the Gauss Manhattan distance and Euclidean distance (63% and 28%, respectively). Experiments on real images reveal that the execution times of the proposed method and the Canny edge detection algorithm slightly increase with image size, as indicated by their execution times of 1.7 and 4.6 s, respectively, for an image size of 1 024 × 768. As the proposed method increases the integral chart and block solution, it achieves a longer execution time than the Canny algorithm. As for the Gauss Manhattan distance and Euclidean distance, their execution times are longer than those of the two previous algorithms. The experiments on simulated and real images prove that the proposed method achieves higher accuracy in edge extraction than other algorithms. This method is also highly practical, as it maintains a relatively short processing time as image size increases.关键词：edge detection;Euclidean distance;Manhattan distance;summed area table;distance map3006|485|1更新时间：2024-05-07
- 摘要：To improve the noise immunity and universality of the fuzzy C-means clustering segmentation algorithm based on a two-dimensional histogram, we propose a weighted fuzzy C-means clustering segmentation method on the basis of a dimensional histogram. The threshold parameter selection inherent in the fuzzy C-means clustering segmentation algorithm based on a two-dimensional histogram leads to poor noise immunity. This issue is addressed in this work with the introduction of weighting properties for the weighted fuzzy C-means clustering segmentation method based on a two-dimensional histogram. This approach is an effective solution for each dimension of the attributes of the poly problem class contribution. Compared with the algorithm based on a two-dimensional histogram, the proposed algorithm shows an average increase of 2 dB to 3 dB in its salt and pepper and Gaussian noise immunity. The same is true for the proposed algorithm when compared with the C-means clustering segmentation algorithm based on fuzzy local information. In the latter comparison, the proposed method reduces its anti-Gaussian noise to less than 1 dB and is 40 times slower than the C-means clustering segmentation algorithm based on fuzzy local information. The proposed method more effectively addresses noisy image segmentation requirements compared with the existing fuzzy C-means clustering algorithm based on a two-dimensional histogram. Moreover, the proposed method is more applicable in target tracking occasions and identification than the fuzzy C-means clustering algorithm based on fuzzy local information.At the same time, a large number of tests proved that the proposed algorithm is suitable for the synthetic images, intelligent traffic images and remote sensing image.关键词：fuzzy C-means clustering;histogram;attribute weighting;image segmentation3249|293|1更新时间：2024-05-07
- 摘要：Automatically segmenting rock CT images is difficult because of the low resolution and fuzzy edges of rock CT images. To achieve desired results, human intervention is necessary in the segmentation process. However, the CT image data for 3D model reconstruction is considerably large. In this case, the workload becomes too heavy to achieve a successful intervention for each frame. In this work, we propose an efficient and practical method that can ensure desired segmentation results with little human intervention. An original threshold is first set. This original threshold comes from one of the CT images used in the study. Then, the threshold for segmenting adjacent frames is calculated automatically using the correlation of the adjacent frames. In this way, all the segmentation objects from the rock CT images are successfully obtained. If necessary, all the CT images are repaired with an automatic region growing method according to the frame with perfect segmentation to improve segmentation. Manual repair is found to be useful when automatic repair is unsuccessful. After the manual repair of some of the objects in one frame, the correlative targets of the other frames are segmented automatically with the proposed method. Improved results are obtained. By testing a series of rock CT images with various scales and resolutions, we find that the method proposed in this work can achieve the desired segmentation at speeds equal to that of the segmentation method based on a fixed threshold and double that of the OSTU method. The proposed segmentation method for rock CT images maximizes the correlation of adjacent frames and can quickly and efficiently extract objects from continuous images in sequence.关键词：histogram;rock;CT images;region growing;threshold segmentation3066|267|1更新时间：2024-05-07
- 摘要：Traditional multiple instance learning (MIL) tracking utilizes self-learning procedures in tracking systems. Once an object gets lost in tracking, the interior classifier easily degenerates. To alleviate this problem, we propose an improved multiple instance learning tracking based on online feature selection (MILOFS). First, a very sparse random matrix is constructed to facilitate the feature initial process. With this matrix, the intrinsic attributes of the features projected from a high-dimension image can be preserved. Then, the loss function of a bag model is built with the Fisher linear discriminant model. The discriminative model of the bag is formed directly at the instance level with the response of each instance. Finally, the gradient descend rule is incorporated into the online boosting framework, and gradient boosting is employed to construct the selection strategy for strong classifiers. Comparison experiments under different scenarios reveal that the center location errors of Online AdaBoost(OAB), Online Multiple Instance Learning Tracking (MIL-Track), Weighted Multiple Instance Learning(WMIL) and Multiple Instance Learning with Online Feature Selection(MILOFS) are 36, 23, 24, and 13 pixels, respectively. Hence, the proposed method is robust and accurate regardless of changes in the illumination, occlusion condition, and target appearance in the outer environment. An improved MILOFS is proposed in this work. The proposed method integrated with a gradient boosting framework and online feature selection strategy effectively addresses the issue of classifier degeneration in traditional MIL tracking.关键词：object tracking;multiple instance learning;Fisher linear discriminant;gradient boosting;discriminant model2873|815|2更新时间：2024-05-07
- 摘要：Object segmentation in video is an important subject in computer vision and has gained various research and application values. The online automatic object segmentation method is proposed in this paper; this method fuses appearance and motion features. First, the object points were roughly estimated by employing appearance and motion boundaries. Then, we utilized these estimated object points to refine the appearance model (GMM) of the previous frame as current appearance model. Second, a Markov random field (MRF) model was constructed by taking the superpixels as nodes, and integrating the appearance model and the location prior. Therefore, the object segmentation can be converted to an energy minimization problem, which is optimized by graph cut in this paper. After extensive experiments which included comparison analysis of five approaches and component analysis of the proposed approach on two datasets, the proposed approach improved accuracy of segmentation by at least 44.8% than other approaches, and achieved higher efficiency of segmentation. The proposed algorithm achieved online automatic object segmentation by fusing appearance and motion features, and obtained good segmentation performance. Furthermore, this algorithm was also robust in several complicated scenes.关键词：feature fusion;Markov random field model;online segmentation;automatic segmentation3159|472|1更新时间：2024-05-07
- 摘要：Three-dimensional reconstruction containing texture information is a classical issue in computer vision. Considering the complexity of an indoor scene and the length of sampling image sequence captured from a random moving RGB-D sensor, conventional three-dimensional reconstruction methods suffer from limited scale and perform poor local detail reconstruction effect. This paper proposes two improvements of the RGBD-SLAM-based three-dimensional reconstruction algorithm to obtain higher quality reconstruction effect. On the one hand, the plane-primitives are incorporated as constraints to enhance robustness and accuracy of the pair-wise registration algorithm. On the other hand, to reduce the influence of RGB-D sensor large distortion, a novel exponential weight function that is motivated by a Gaussian noise model is proposed. In the experiment, the proposed method yields higher quality results compared with state-of-the-art approaches on the benchmarks dataset of the computer vision group of Stanford. Our method also achieves lower average absolute trajectory error compared with a conventional RGB-D SLAM method. Experimental results demonstrate that our method substantially increases the accuracy of camera pose estimation and quality of indoor scene three-dimensional reconstruction.关键词：RGB-D sensor;simultaneous localization and mapping(SLAM);camera trajectory estimation;3D reconstruction4375|393|7更新时间：2024-05-07
- 摘要：With the rapid development of science and technology, geometric modeling techniques for curves have recently gained significant research interest. According to the 2D and 3D freeform deformation of parametric curves and free curves, a new algorithm based on polynomial extension function with platform is proposedto obtain several deformation effects. The new deformation method possesses perfect properties such as interval peak, peak value and symmetries. Furthermore, it has the closure property for deformations of Bézier and NURBS curves which are the mainFstream curves of present modeling system. A simple extension function is first presented. Unlike most existing freeform deformation methods, the proposed extension function has an ordinary polynomial form. Hence, this extension function does not include transcendental functions and complex functions. With this extension function, one new extension factor with extension and smoothing parameters is also established. This new extension factor constructed with the extension function possesses perfect properties, including interval peaks, peak values, and symmetries. Finally, the extension matrixes are applied to the parametric and freeform curves to obtain the global, local, periodic, and elastic deformation effects. This objective is achieved by adjusting the deformation intervals, extension parameters, smoothing parameters, and deformation directions. The deformation method remains self-enclosed when implemented on Bezier and NURBS curves, which are mainstream curves in CAGD. Several numerical examples show that the proposed method is easy to calculate and control and that it can be used repeatedly to achieve a large number of contour lines. Compared with many freeform deformation methods, the proposed algorithm can be used for not only general 2D and 3D parametric curves but also freeform curves. Thus, it demonstrates extensive application scope. Because the extension function has perfect properties such as interval peak, peak value and symmetries, it can produce different kinds of special curves with angular points and cuspidal points which the other deformation method can't achieve. In sum, the proposed algorithm greatly enriches the deformation effects of curves.关键词：free curve;platform extension function;interval peak;freeform deformation2450|233|3更新时间：2024-05-07
- 摘要：To address the problem of how to produce variform individual texture images of tree from the same tree species through the method of “bulletin board” mapping in scene modeling, a method to generate tree texture images through the use of a random transmogrification mesh is proposed. The original mesh is utilized to divide tree texture images. Then, random transmogrification of the mesh vertexes is conducted. Lastly, the generated irregular mesh images are mapped on a corresponding regular mesh in the new images; texture images are obtained after transmogrification. The effects of mesh division quantity, transformative coefficient of the mesh, and quantity of mapping nodes on the transformative results are simulated. By selecting the mesh division number, transformative coefficient of the mesh, and quantity of mapping nodes appropriately, the original texture images of trees after transmogrification can be converted into multiple ones that have similar basic forms but different details. Experimental results indicate clearly that the problem of texture image generation of different individual tree from the same species can be solved by the proposed method. The proposed method involves a stochastic parameter to control the local transmogrification extent of tree images; the environmental impact on the individual form exerted by various random factors in their growing environment is reflected. The method is much easier and more convenient than individual tree image mapping. Additionally, the visual effects produced by this method are more real than those produced by mapping multiple individuals in the same image. A new method to generate tree texture images through the use of a random transmogrification mesh is proposed. The transmogrification extent of tree images is controlled by a stochastic parameter; the effects on the individual forms exerted by various random factors in their own growing environment are simulated. Images of individuals with various shapes in the same species can be obtained by selecting the appropriate transformative parameters. Experimental results show that when the tree texture image produced by the proposed method is applied in “bulletin board” mapping, the scene of multiple individuals of the same species can be simulated well in scene modeling. Local transmogrification of the image shows that the continuity of the original image content remains, including the overall basic shape and internal structure. Therefore, this method is also applicable to image processing of several other out-of-shape natural scenes (e.g., mountains, rivers, ocean waves).关键词：scene modeling;tree texture image;random transmogrification;tree form2797|243|2更新时间：2024-05-07
- 摘要：Seldom does a photograph record what we perceive with our eyes. As a result, we could not obtain an optimal photo with all the subjects smiling as expected. When taking pictures, we spend a considerable amount of time capturing other shots and inadvertently miss beautiful scenes. Traditional photo editing apps could not facilitate expression editing. Hence, we develop an algorithm for interactive photo editing based on expression transmission to edit parts of faces with poor expression. The proposed algorithm uses scan line deformation or feature point deformation on the basis of the identity related to the source face. Our approach is to utilize target face expression to edit source faces with unsatisfied expressions. The result shares an expression similar to that of the target face. The face is edited by relighting and then finally fused in the photo. The proposed algorithm can edit frontal face images to achieve desired expressions. This algorithm, which is based on different processing conditions and transmission models, demonstrate better performance than the conventional Gaussian mixture model with regard to the test results. This study proposes a novel expression transmission model. We assign different transmission models for different pieces of target face identity information and subsequently introduce interaction. Experiment results indicate that our model achieves preferable adaptive performance when applied in photos with different conditions and requirements.关键词：expression mapping;feature points detection;relighting;Poisson image fusion2506|239|0更新时间：2024-05-07
- 摘要：Snakes models can effectively fit curve features and are thus widely used to extract roads from remote sensing images. However, when used to extract roads from synthetic aperture radar (SAR) images, traditional Snakes models that utilize the negative gradient of images as external energy cannot obtain the desired results because of serious speckle noise. To address this issue, we employed a tensor voting method in improving Snakes models because such method can extract salient structures from images influenced by noises. Road class was first segmented from SAR images using the FCM clustering method. Then, the saliency value of the curve features was obtained by employing the tensor voting method on the extracted road class. Finally, the negative normalized saliency value of the curve features was used as the external energy of the snakes model to extract roads. To minimize the energy of the snakes model, a strategy for minimizing energy while interpolating nodes was proposed. Road extraction experiments were performed on different scenes of spaceborne and airborne SAR images. Compared with a similar method based on the snakes model, the proposed modelachieved better fitting results with less control points. Moreover, the proposed method showed better detection completeness, correctness, and quality than the MRF-based method. The proposed method also demonstrated a shorter detection time, which is a practical feature for wide-range road network extraction. The proposed method quantified the geometric characteristics of roads through tensor voting. An optimized fitting strategy was used to minimize the energy consumption of the snakes model for road extraction. The experiments on spaceborne and airborne SAR images proved that main roads in rural and urban scenes can be effectively extracted using the proposed method.关键词：tensor voting;computer vision algorithm;saliency;snakes model;SAR images;road extraction3774|307|4更新时间：2024-05-07
- 摘要：Existing methods for visualizing global-scale terrains are basically derived from the chunked LOD method. With such methods, GPU-CPU data transfer bottlenecks still emerge in real-time walkthroughs. Moreover, the addition of skirts to fix cracks requires additional resources and may not even eliminate artifacts completely.To address these issues, we proposed a GPU terrain visualization method that combines the GPU tessellation algorithm. The view-dependent and screen space error-controlled LOD method, and local coordinate system rendering algorithms. This method significantly improved the efficiency of global terrain visualization. A global terrain visualization system was implemented to provide global multi-resolution satellite images and achieve high-resolution terrain field data rendering. Extensive experiments were performed to analyze the system benchmark. Compared with the traditional GPU-based chunked LOD method, the proposed method can improve FPS by more than 100%, thus eliminating the system bottleneck problem. The proposed method is practicable, robust, and widely applicable in large-scale global rendering systems.关键词：global visualization;terrain rendering;GPU rendering;GPU Tessellation;dynamic local coordinate3252|199|3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