最新刊期
卷 17 ， 期 12 ， 2012
- 摘要：Gaussian Mixture Models(GMMs) is the basic model of statistical machine learning and widely applied to visual media fields. In recently years, with the rapid growth of visual media information and deep development of analytical techniques GMMs have obtained further developments in such fields as (texture) image segmentation, video analysis, image registration and clustering. This paper begins from the basic models of GMMs, discusses and analyzes from both theoretical and application aspects the solving methods of GMMs including EM algorithms and its variants, and expounds the two problems of model selection: online learning and model reduction. In visual applications, this paper introduces GMM-based models and methods in image segmentation, video analysis, image registration and image de-noising, expatiates the principles and processes of some newest and classical models, such as space-variant GMMs for image segmentation, coherent point draft algorithm for image registration. At last, this paper gives some possible latent directions and difficult problems.关键词：Gaussian mixture models(GMMs);EM algorithm;clustering analysis;image segmentation;object recognition;image registration;vision4570|306|0更新时间：2024-05-08
- 摘要：Gaussian noise in color images is difficult to remove. We propose a denoising method by combining chrominance median filter based on chrominance model and Bayes multi-threshold filter in the curvelet transform domain. The method converts RGB images to HSI images first. The noise in the H and S components is filtered by searching chrominance median in the chrominance plane. Then, the noise in the I component is removed by using Bayes multi-threshold filter in the cycle spinning curvelet domain. Finally, the denoised color images are produced through the HSI inverse transformation. The contrast experiment results indicate that while achieving higher PSNR, the proposed method can better preserve the color images’ chrominance, intensity as well as texture information, while avoiding yielding artifact effect for Gaussian noise removal in color images.关键词：HSI space;image denosing;chrominance model;curvelet transform;Gaussian noise3144|117|0更新时间：2024-05-08
- 摘要：The incremental dissimilarity approximations (IDA) algorithm is a recently proposed high-efficient fast image pattern matching algorithm. By splitting the matching vectors, the IDA algorithm saves a lot of pixel-dependment calculations. However,the sub-vectors have a rather weak energy compaction after splitting. This means IDA’s efficiency can further be improved. To avoid the weak energy compaction, sub-vector ordering is proposed, which sorts the sub-vectors by their variances. Candidates would be pruned earlier by the sorted order in pattern matching. Therefore, the average number of unfolded sub-vectors is reduced, which also reducts the search space. Additionally,one more pruning test using the whole vector’s norm before IDA is proposed in our work, and the PDS (partial distortion search) algorithm is introduced in the unfolding sub-vectors step. In our experiment, by testing three types of images in the data sets(indoor scene, natural scene, streetscape), the overall efficiency of proposed algorithm is improved by 72%~83% compared to the original IDA algorithm.关键词：fast pattern matching;vector partitioning;IDA algorithm;variance sorting;high resolution image2535|37|0更新时间：2024-05-08
- 摘要：The traditional Lucy-Richardson algorithm is an iterative image restoration method based on Bayesian analysis. It achieves good results for restoring images degraded with a high signal-to-noise ratio(SNR). The algorithm is so sensitive to noises that some regularized methods are introduced into the LR-algorithm. However, these tricks often tend to produce excessive smoothing. Therefore, in this paper,we introduce the image sparse prior model as a regularization item into the LR-algorithm, and get a new regularization LR algorithm to suppress noise amplification in the iterative process. To be different from the conventional gratitude-restriction approaches, the algorithm proposes a varying parameterized sparse gradient regularization restriction method, which enables the gradient distribution parameters of the restored image more close to the true gradient distribution and avoids excessive smoothing of restored image by adjusting the regular coefficient. The experimental results show that the algorithm can efficiently suppress the amplification of noises and preserve the details of images.关键词：image restoration;Lucy-Richardson algorithm;image sparse gratitude distribution;regularization3949|91|0更新时间：2024-05-08
- 摘要：Motion estimation is significant in the applications of video image compression, super-resolution reconstruction, mosaic, and target detection, and so on. In this letter, we present an improved algorithm for the problem of highly accurate sub-pixel global motion estimation, which introduces an optimal filter for computing image gradient, and also adopts the method of combining upsampling with paraboloid fitting gradient cross-correlation. Experimental results show that the proposed algorithm can not only achieve good robustness to the influence of noise, but can also improve the accuracy of motion estimation significantly.关键词：sub-pixel;global motion estimation;optimal filter;gradient cross-correlation;upsampling;matrix multiplication;paraboloid fitting3191|44|0更新时间：2024-05-08
- 摘要：An improved adaptive Chord-to-Point distance accumulation algorithm (ACPDA) is proposed which can handle adjacent corners, obtuse corners, and round corners. The main advantages of our implementation include: 1) the neighborhood of the candidate corners is re-detected for avoiding loss of the adjacent corners, which making the localization of corners more precise; 2) an adaptive threshold for each curve is set for removing the false corners, avoiding loss of the obtuse corner; 3) a local adaptive threshold is constructed to remove the round corners effectively. The experimental results have evaluated the performance of the ACPDA.关键词：corner detection;adaptive chord-to-point distance accumulation;adjacent corner;adaptive threshold3167|83|0更新时间：2024-05-08
- 摘要：In order to deal with the kernel optimization, a new kernel data-dependent optimizaition kernel approach based on maximum subclass margin criterion is proposed. In this scheme, a maximum subclass margin function is created firstly. Then, the in-between-subclass and inter-subclass scatter matrix in the empirical feature space are defined. Finally, the optimal coefficients vector is solved by the selected optimization criterion. Experimental results based on UCI data show that it is effective and feasible.关键词：Kernel function;kernel optimization;maximum subclass margin criterion;target recognition2766|46|0更新时间：2024-05-08
- 摘要：In this study,we propose a new kernel discriminant for learning and recognition of image sets using canonical correlation. Each image set is mapped into a high-dimensional feature space. The corresponding kernel space is then constructed by a kernel linear discriminant analysis.The similarity of two kernel subspaces is assessed by calculating the canonical difference between them. According to the kernel Fisher discriminant, a Kernel Discriminant Analysis of Canonical Correlation algorithm is derived to establish the correlation between the kernel subspaces based on the ratio of the canonical differences of the between-classes to those of the within-classes. The experimental results on the ORL, NUST603, FERNT and XM2VTS database demonstrate that the proposed method can efficiently extract the features of the images. Moreover, the recognition rate of the proposed algorithm outperforms DCC and KDT.关键词：canonical correlation;canonical difference;kernel linear discriminant analysis;kernel discriminant transformation;face recognition3106|47|0更新时间：2024-05-08
- 摘要：Obtain appropriate low-dimension face features is an important problem in the area of face recognition. Traditional face recognition algorithms based on wavelet transform extract image features using only the low frequency components for classification, which results in the loss of information,which could be used for face recognition. In order to effectively extract the face image features, a new algorithm of face recognition based on wavelet transform and weighted fusion of features is proposed in this study. First, the wavelet transform is used to reduce the dimensionality; then,the features of the four wavelet sub-graphs are extracted by a principal component analysis (PCA), and the features of the four parts are fused into discriminant features. Finally, the features are classified and recognized by SVM. Experimental results on the ORL face database show that the proposed algorithm achieves a recognition accuracy of 97.5 percent, so the new algorithm can effectively improve the face recognition ability. It has a higher recognition accuracy than traditional methods.关键词：face recognition;wavelet transform;principal component analysis;weighted fusion;support vector machine3456|162|0更新时间：2024-05-08
- 摘要：In order to solve the problem of target identification label confusion in multi-target tracking,especially with occluded or stacked targets, an algorithm based on color features is proposed for multi-target tracking. In the process of target tracking, the foreground is derived using the background subtraction method. The blobs in the foreground are classified into noise regions, single target regions,and multi-target regions. According to the blob classification, a different processing mechanism is used. The system adapts the correction time stamp to process noise regions, using Kalman prediction processing for fast motions and using the mean shift algorithm for processing target identification labels.Through several experiments, we show that the new algorithm has a good real-time performance (the tracking speed is 30 f/s), has a very strong background suppression, and has the characteristics for long-time target tracking.关键词：real-time detection;background modeling;mean shift;identification Label2691|117|0更新时间：2024-05-08
- 摘要：A curved surface marker, which can be used for cone pose measurements, is designed. The design principles of the curved surface marker are proposed. The cone is flattened into a circular sector and then the sector is divided into six equal sub-sectors, which are used as basic design units. Trapezoids are applied to fulfill the joint design of point features and area features, and to finish the form-giving design of the marker, guaranteeing that there are 58 corners in each sub-sector. Combining the geometric features with the movement features of the cone, a unique code is provided for each sub-sector. The simulation results show that more than five non-fully collinear corners can be extracted and recognized under various poses (with the cone top appears in FOV of both cameras simultaneously). The designed marker can satisfactory the requirements of cone pose measurement.关键词：cone;pose measurement;marker;corner2852|39|0更新时间：2024-05-08
- 摘要：Considering the work needed for constructing muscle models artificially,setting their control nodes, and adjusting their computer parameters, we present a method to construct the muscle model automatically and to generate the model calculation parameters for 3D facial animation. We developed a robust facial features recognition algorithm to extract the geometry and texture feature vertices. In the geometry feature recognition process, we adopt synthetically several constraints related to the Gaussian curvature and surface normal value to extract the candidate vertices. In the texture feature recognition process, we use the Gaussian Mixture Model of CrCgCb to extract the feature vertices. Then, clustering procedures are applied to gain the final feature vertices. Finally,using the 13 geometry feature vertices and 8 texture feature vertices extracted by the recognition algorithm,we automatically construct the muscle models for the real-time facial animation.The experimental results demonstrate a matching rate over 90% compared with the landmark vertices made by an artist. The application work indicates that the process of automated muscle model construction based on the feature recognition algorithm fit in with different human head geometries very well. On this basis, we synthesize a group of characteristic facial expressions and mouth shapes with higher realism in real time.关键词：facial animation;facial feature recognition;automatic animal modeling;automatic vertex tagging3109|81|0更新时间：2024-05-08
- 摘要：Virtual reality is the most important character of 3D GIS, whose essence is the integration of visualization technology and GIS databases to meet the needs of a variety of applications such as agricultureor, disaster prediction. In this study, based on the environmental data and meteorological data of a GIS database, meteorological data can be visualized in near real-time, through the simulation of rain and snow effects. In this paper, we use particle systems to reach better real-time effect with lower memory consumption through endowing every particle with a large area texture. Thus, the number of particles reduces to 10% of common particle system and rendering speed rises over 10 times. GPU coding is used to accelerate 3D rendering of the ground effect of the rain. The prototype system can accept the settings of particle texture and rainfall or snowfall intensity by users. To satisfy practical applications, we propose a self-adaptive strategy based on meteoro-logical data.关键词：3D geographical information system(3D GIS);snow and rain simulation;natural phenomena;real-time rendering3134|66|0更新时间：2024-05-08
-
Fuzzy C-means clustering algorithm combined with markov random field for brain MR image segmentation
摘要：Brain magnetic resonance imaging (MRI) has been widely used in clinical practice.Accurate segmentation of brain tissue structure can improve the reliability of the brain disease diagnosis and the effectiveness of treatments. The fuzzy C-Means Clustering (FCM) algorithm is good at solving ambiguities and uncertainties in images, and it is one of the most common brain MRI segmentations. However, FCM has a poor anti-noise ability, because it only uses the grayscale information without considering regional information. The Markov Random Field (MRF) algorithm takes full advantage of the image regional information, but it tends to over-segment. Therefore, we use FCM often combined with MRF to improve the results. In this paper, considering the problem in the existing combination algorithms of FCM and MRF, we propose a new adaptive weight combination of FCM and MRF algorithm for brain MRI segmentation. The algorithm adaptively updates the combining field weight parameter , using spatial relativity of the adjacent pixel regions. It improves the existing fixed weight combination methods of FCM and MRF, and makes full use of FCM and MRF. Experiment results show that this algorithm has stronger anti-noise property and higher segmentation precision than FCM and some other FCM improved algorithms.关键词：fuzzy C-means clustering;Markov random field;MR image;image segmentation;regional information3233|130|0更新时间：2024-05-08 - 摘要：Due to the variation of imaging conditions onboard, the chromaticity among a collection of remote sensing images that are to create image mosaics often differs. In this regard, a method for keeping color consistency based on support vector regression (SVR) is presented. First, the pixels, with invariant features, are automatically selected in the overlapping areas, which are based on the image threshold segmentation and on spectral angle matching (SAM). These pixels are used to build transformation equations on digital numbers between the original image and the reference image using SVR. Finally, the brightness of the images to be stitched is corrected to the same reference image using the corresponding transformation equations. The approach mentioned was implemented on the thematic mapper (TM) imagery, SPOT satellite imagery and unmanned aerial vehicle (UAV) images. Our results show that this method can effectively alleviate the chromaticity differences. Compared with the linear regression method, the above method achieves larger variance and higher radiation resolution.关键词：remote sensing image mosaic;color normalization;spectral angle matching;support vector regression3616|55|0更新时间：2024-05-08
- 摘要：We introduce the projection pursuit regression (PPR)analysis for remote sensing image classification and describe the implementation of the PPR model used in remote sensing image classification. Using a TM image of the Guangzhou area for our classification tests,we get satisfactory classification result, after optimizing the parameters in the projection pursuit regression model by shuffled frog leaping algorithm. Furthermore, we discuss the setting of the projection center as well as the influence of the optimal algorithms and the number of ridge functions on the classification accuracy in the PPR model. The results show that the model is easy to realize and very stable.The number of ridge functions in the projection pursuit regression model has no significant influence on the classification accuracy.关键词：image classification;projection pursuit regression;ridge function;projection center;optimization algorithm2853|255|0更新时间：2024-05-08
- 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