Catálogo de publicaciones - libros

Based on the equivalence between canonical correlation analysis (CCA) and Fisher linear discriminant analysis (FLDA), two methods for feature extraction of face images are proposed in this paper. In the first approach, the high-dimensional face images are first mapped into the range space of total scatter matrix using principle component analysis (PCA). Then CCA is performed to extract the linear optimal discriminant features without losing Fisher discriminatory information. In the second approach, nonlinear features are extracted using KPCA+CCA which is equivalent to KFDA in nature. The experimental results upon ORL face database indicate that the proposed PCA/KPCA+CCA significantly outperform the traditional Fisherface method.

Fuzzy Kohonen clustering network(FKCN) shows great superiority in processing the clustering in image segmentation. In this paper, an intensified Fuzzy Kohonen clustering Network (IFKCN) is proposed for texture segmentation. The method adjusts fuzzy factors to accelerate the speed of convergence. It intensifies the biggest membership and suppresses the other. By using this network in Brodatz texture segmentation, its iteration is fewer and the speed of convergence is quicker than FKCN and AFKCN(Adaptive Fuzzy Kononen clustering Network), and segmentation results are as well as FKCN.

Support Vector Machine (SVM) is a very effective method for pattern recognition. In this article, a intelligent diagnosis system based on SVMs is presented to solve the problem that there is not effective method for reciprocating compressor valve fault detection. The Local Wave method was used to decompose vibration signals, which acquired from valves surface, into sub-band signals. Then the higher-order statistics were calculated as the input features of classification system. The experiment results confirm that the classification technique has high flexibility and reliability on valve condition monitoring.

In this paper, we introduce a system that recognize emotion by speech and show the facial expression by using 2-dimensional emotion space. 4 emotional states are classified by the work with ANN. The derived features of the signal, pitch, and loudness are quantitatively contributed to the classification of emotions. Firstly we analyze the acoustic elements for using as emotional features and the elements are evaluated by ANN classifier. Secondly, we implement an avatar (simply drawn face) and the facial expressions are changed naturally by using the dynamic emotion space model.

This paper proposes a novel intrusion detection approach by applying kernel principal component analysis (KPCA) for intrusion feature extraction and followed by support vector machine (SVM) for classification. The MIT’s KDD Cup 99 dataset is used to evaluate these feature extraction methods, and classification performances achieved by SVM with PCA and KPCA feature extraction are compared with those obtained by PCR and KPCR classification methods and by SVM without application of feature extraction. The results clearly demonstrate that feature extraction can greatly reduce the dimension of input space without degrading the classifiers’ performance. Among these methods, the best performance is achieved by SVM using only four principal components extracted by KPCA.

Independent component analysis (ICA) is a feature extraction technique for blind source separation. Enhanced independent component analysis (EICA), which has enhanced generalization performance, operates in a reduced principal component analysis (PCA) space. SVM is a powerful supervised learning algorithm, which is rooted in statistical learning theory. SVM has demonstrated high generalization capabilities in many pattern recognition problems. In this paper, we integrate EICA with SVM and apply this new method to the leak detection problem in oil pipelines. In features extraction, EICA produces EICA features of the original pressure images. In classification, SVM classified the EICA features as leak or non-leak. The test results based on real data indicate that the method can detect many leak faults from a pressure curve, and reduce the ratio of false and missing alarm than conventional methods.

Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) techniques are important and well-developed area of image recognition and to date many linear discrimination methods have been put forward. Despite these efforts, there persist in the traditional PCA and LDA some weaknesses. In this paper, we propose a new Line-based methodes called Line-based PCA and Line-based LDA that can outperform the traditional PCA and LDA methods. As opposed to conventional PCA and LDA, those new approaches are based on 2D matrices rather than 1D vectors. That is, we firstly divide the original image into blocks. Then, we transform the image into a vector of blocks. By using row vector to represent each block, we can get the new matrix which is the representation of the image. Finally PCA and LDA can be applied directly on these matrices. In contrast to the covariance matrices of traditional PCA and LDA approaches, the size of the image covariance matrices using new approaches are much smaller. As a result, those new approaches have three important advantages over traditional ones. First, it is easier to evaluate the covariance matrix accurately. Second, less time is required to determine the corresponding eigenvectors. And finally, block size could be changed to get the best results. Experiment results show our method achieves better performance in comparison with the other methods.

Transportation system is a complex, large, integrated and open system. It’s difficult to recognize the system with analytical methods. So, two neural network models are developed to recognize the system. One is a back propagation neural network to recognize ideal system under equilibrium status, and the other is a counter propagation model to recognize real system with probe vehicle data. By recognizing ideal system, it turn out that neural network can simulate the process of traffic assignment, that is, neural network can simulate mapping relationship between OD matrix and assigned link flows, or link travel times. Similarly, if real-time OD matrix is obtained by probe vehicle technology, and then similarly results like link travel times can be obtained by similarly models. By comparing outputs of two models, difference about real and ideal transportation system can be found.

In this paper we propose a method of locating face and eyes using context-aware binarization. Face detection obtains the face region using neural network and mosaic image representation. Eye location extracts the location of eyes from the detected face region. The proposed method is composed of binarization, connected region segmentation by labeling, eye candidate area extraction by heuristic rules that use geometric information, eye candidate pair detection, and eye area pair determining by ranking method. Binarization plays an important role in this system that converts a source image to a binary image suitable for locating eyes. We consider edge detection based and image segmentation based binarization methods. However, each method alone cannot be used a solution in general environment because these are influenced by the factors such as light direction, contrast, brightness, and spectral composition. We propose a hybrid binarization using the concept of illumination context–awareness that mixes two binarization methods in general environment.

This paper presents a face recognition algorithm by using Gabor wavelet transform for facial features extraction and Support Vector Machines (SVM) for face recognition, Gabor wavelets coefficients are used to represent local facial features. The implementations of our algorithm are as follows: Firstly, facial feature points are located roughly by using a set of node templates. Secondly, Gabor wavelet coefficients are extracted at every facial feature point, and all the Gabor wavelet coefficients are catenated to represent a face image. Lastly, SVM classifiers are used for face recognition. The experimental results demonstrate the effectiveness of our face recognition algorithm.