When dealing with pattern recognition problems one encounters different types of prior knowledge. It is important to incorporate such knowledge into classification method at hand. A very common type of prior knowledge...When dealing with pattern recognition problems one encounters different types of prior knowledge. It is important to incorporate such knowledge into classification method at hand. A very common type of prior knowledge is many data sets are on some kinds of manifolds. Distance based classification methods can make use of this by a modified distance measure called geodesic distance.We introduce a new kind of kernels for support vector machines which incorporate geodesic distance and therefore are applicable in cases such transformation invariance is known. Experiments results show that the performance of our method is comparable to that of other state-of-the-art method.展开更多
It has been demonstrated that the linear discriminant analysis (LDA) is an effective approach in face recognition tasks. However, due to the high dimensionality of an image space, many LDA based approaches first use t...It has been demonstrated that the linear discriminant analysis (LDA) is an effective approach in face recognition tasks. However, due to the high dimensionality of an image space, many LDA based approaches first use the principal component analysis (PCA) to project an image into a lower dimensional space, then perform the LDA transform to extract discriminant feature. But some useful discriminant information to the following LDA transform will be lost in the PCA step. To overcome these defects, a face recognition method based on the discrete cosine transform (DCT) and the LDA is proposed. First the DCT is used to achieve dimension reduction, then LDA transform is performed on the lower space to extract features. Two face databases are used to test our method and the correct recognition rates of 97.5% and 96.0% are obtained respectively. The performance of the proposed method is compared with that of the PCA+ LDA method and the results show that the method proposed outperforms the PCA+ LDA method.展开更多
In order to solve the model update problem in mean-shift based tracker, a novel mechanism is proposed. Kalman filter is employed to update object model by filtering object kernel-histogram using previous model and cur...In order to solve the model update problem in mean-shift based tracker, a novel mechanism is proposed. Kalman filter is employed to update object model by filtering object kernel-histogram using previous model and current candidate. A self-tuning method is used for adaptively adjust all the parameters of the filters under the analysis of the filtering residuals. In addition, hypothesis testing servers as the criterion for determining whether to accept filtering result. Therefore, the tracker has the ability to handle occlusion so as to avoid over-update. The experimental results show that our method can not only keep up with the object appearance and scale changes but also be robust to occlusion.展开更多
A novel wavelet-based algorithm for image enhancement is proposed in the paper. On the basis of multiscale analysis, the proposed algorithm solves efficiently the problem of noise over-enhancement, which commonly occu...A novel wavelet-based algorithm for image enhancement is proposed in the paper. On the basis of multiscale analysis, the proposed algorithm solves efficiently the problem of noise over-enhancement, which commonly occurs in the traditional methods for contrast enhancement. The decomposed coefficients at same scales are processed by a nonlinear method, and the coefficients at different scales are enhanced in different degree. During the procedure, the method takes full advantage of the properties of Human visual system so as to achieve better performance. The simulations demonstrate that these characters of the proposed approach enable it to fully enhance the content in images, to efficiently alleviate the enhancement of noise and to achieve much better enhancement effect than the traditional approaches. Key words wavelet transform - image contrast enhancement - multiscale analysis CLC number TP 391 Foundation item: Supported by the National Natural Science Foundation of China (69931010)Biography: Wu Ying-qian (1974-), male, Ph. D, research direction: image processing, image compression and wavelet.展开更多
Different modalities in biomedical images, like CT, MRI and PET scanners, provide detailed cross-sectional views of human anatomy. This paper introduces three-dimensional brain reconstruction based on CT slices. It co...Different modalities in biomedical images, like CT, MRI and PET scanners, provide detailed cross-sectional views of human anatomy. This paper introduces three-dimensional brain reconstruction based on CT slices. It contains filtering, fuzzy segmentation, matching method of contours, cell array structure and image animation. Experimental results have shown its validity. The innovation is matching method of contours and fuzzy segmentation algorithm of CT slices.展开更多
This paper explores brain CT slices segmentation technique and some related problems, including contours segmentation algorithms, edge detector, algorithm evaluation and experimental results. This article describes a ...This paper explores brain CT slices segmentation technique and some related problems, including contours segmentation algorithms, edge detector, algorithm evaluation and experimental results. This article describes a method for contour-based segmentation of anatomical structures in 3D medical data sets. With this method, the user manually traces one or more 2D contours of an anatomical structure of interest on parallel planes arbitrarily cutting the data set. The experimental results showes the segmentation based on 3D brain volume and 2D CT slices. The main creative contributions in this paper are: (1) contours segmentation algorithm; (2) edge detector; (3) algorithm evaluation.展开更多
Blind separation of independent sources from their nonlinear convoluted mixtures is a more realistic problem than from linear ones. A solution to this problem based on the Entropy Maximization principle is presented. ...Blind separation of independent sources from their nonlinear convoluted mixtures is a more realistic problem than from linear ones. A solution to this problem based on the Entropy Maximization principle is presented. First we propose a novel two-layer network as the de-mixing system to separate sources in nonlinear convolved mixture. In output layer of our network we use feedback network architecture to cope with convoluted mixtures. Then we derive learning algorithms for the two-layer network by maximizing the information entropy. Based on the comparison of the computer simulation results, it can be concluded that the proposed algorithm has a better nonlinear convolved blind signal separation effect than the H.H. Y's algorithm.展开更多
Object-based video segmentation is an important issue for many multimedia applications. A video segmentation method based on EM algorithm is proposed. We consider video segmentation as an unsupervised classification p...Object-based video segmentation is an important issue for many multimedia applications. A video segmentation method based on EM algorithm is proposed. We consider video segmentation as an unsupervised classification problem and apply EM algorithm to obtain the maximum-likelihood estimation of the Gaussian model parameters for model-based segmentation. We simultaneously combine multiple features (motion, color) within a maximum likelihood framework to obtain accurate segment results. We also use the temporal consistency among video frames to improve the speed of EM algorithm. Experimental results on typical MPEG-4 sequences and real scene sequences show that our method has an attractive accuracy and robustness.展开更多
基金Supported by National Natural Science Foundation of P. R. China (50174038, 30170274)
文摘When dealing with pattern recognition problems one encounters different types of prior knowledge. It is important to incorporate such knowledge into classification method at hand. A very common type of prior knowledge is many data sets are on some kinds of manifolds. Distance based classification methods can make use of this by a modified distance measure called geodesic distance.We introduce a new kind of kernels for support vector machines which incorporate geodesic distance and therefore are applicable in cases such transformation invariance is known. Experiments results show that the performance of our method is comparable to that of other state-of-the-art method.
文摘It has been demonstrated that the linear discriminant analysis (LDA) is an effective approach in face recognition tasks. However, due to the high dimensionality of an image space, many LDA based approaches first use the principal component analysis (PCA) to project an image into a lower dimensional space, then perform the LDA transform to extract discriminant feature. But some useful discriminant information to the following LDA transform will be lost in the PCA step. To overcome these defects, a face recognition method based on the discrete cosine transform (DCT) and the LDA is proposed. First the DCT is used to achieve dimension reduction, then LDA transform is performed on the lower space to extract features. Two face databases are used to test our method and the correct recognition rates of 97.5% and 96.0% are obtained respectively. The performance of the proposed method is compared with that of the PCA+ LDA method and the results show that the method proposed outperforms the PCA+ LDA method.
文摘In order to solve the model update problem in mean-shift based tracker, a novel mechanism is proposed. Kalman filter is employed to update object model by filtering object kernel-histogram using previous model and current candidate. A self-tuning method is used for adaptively adjust all the parameters of the filters under the analysis of the filtering residuals. In addition, hypothesis testing servers as the criterion for determining whether to accept filtering result. Therefore, the tracker has the ability to handle occlusion so as to avoid over-update. The experimental results show that our method can not only keep up with the object appearance and scale changes but also be robust to occlusion.
文摘A novel wavelet-based algorithm for image enhancement is proposed in the paper. On the basis of multiscale analysis, the proposed algorithm solves efficiently the problem of noise over-enhancement, which commonly occurs in the traditional methods for contrast enhancement. The decomposed coefficients at same scales are processed by a nonlinear method, and the coefficients at different scales are enhanced in different degree. During the procedure, the method takes full advantage of the properties of Human visual system so as to achieve better performance. The simulations demonstrate that these characters of the proposed approach enable it to fully enhance the content in images, to efficiently alleviate the enhancement of noise and to achieve much better enhancement effect than the traditional approaches. Key words wavelet transform - image contrast enhancement - multiscale analysis CLC number TP 391 Foundation item: Supported by the National Natural Science Foundation of China (69931010)Biography: Wu Ying-qian (1974-), male, Ph. D, research direction: image processing, image compression and wavelet.
基金This project was supported by the National Natural Science Foundation of China (69931010).
文摘Different modalities in biomedical images, like CT, MRI and PET scanners, provide detailed cross-sectional views of human anatomy. This paper introduces three-dimensional brain reconstruction based on CT slices. It contains filtering, fuzzy segmentation, matching method of contours, cell array structure and image animation. Experimental results have shown its validity. The innovation is matching method of contours and fuzzy segmentation algorithm of CT slices.
文摘This paper explores brain CT slices segmentation technique and some related problems, including contours segmentation algorithms, edge detector, algorithm evaluation and experimental results. This article describes a method for contour-based segmentation of anatomical structures in 3D medical data sets. With this method, the user manually traces one or more 2D contours of an anatomical structure of interest on parallel planes arbitrarily cutting the data set. The experimental results showes the segmentation based on 3D brain volume and 2D CT slices. The main creative contributions in this paper are: (1) contours segmentation algorithm; (2) edge detector; (3) algorithm evaluation.
文摘Blind separation of independent sources from their nonlinear convoluted mixtures is a more realistic problem than from linear ones. A solution to this problem based on the Entropy Maximization principle is presented. First we propose a novel two-layer network as the de-mixing system to separate sources in nonlinear convolved mixture. In output layer of our network we use feedback network architecture to cope with convoluted mixtures. Then we derive learning algorithms for the two-layer network by maximizing the information entropy. Based on the comparison of the computer simulation results, it can be concluded that the proposed algorithm has a better nonlinear convolved blind signal separation effect than the H.H. Y's algorithm.
文摘Object-based video segmentation is an important issue for many multimedia applications. A video segmentation method based on EM algorithm is proposed. We consider video segmentation as an unsupervised classification problem and apply EM algorithm to obtain the maximum-likelihood estimation of the Gaussian model parameters for model-based segmentation. We simultaneously combine multiple features (motion, color) within a maximum likelihood framework to obtain accurate segment results. We also use the temporal consistency among video frames to improve the speed of EM algorithm. Experimental results on typical MPEG-4 sequences and real scene sequences show that our method has an attractive accuracy and robustness.
