Catálogo de publicaciones - libros

This paper introduces a morphological neural network approach to extract vehicle targets from high resolution panchromatic satellite imagery. In the approach, the morphological shared-weight neural network (MSNN) is used to classify image pixels on roads into vehicle targets and non-vehicle targets, and a morphological preprocessing algorithm is developed to identify candidate vehicle pixels. Experiments on 0.6 meter resolution QuickBird panchromatic data are reported in this paper. The experimental results show that the MSNN has a good detection performance.

In this paper, we propose a secure authentication method based on multimodal biometrics system under ubiquitous computing environments. For this, the face and signature images are acquired in PDA and then each image with user ID and name is transmitted via WLAN (Wireless LAN) to the server and finally the PDA receives authentication result from the server. In the proposed system, face recognition algorithm is designed by PCA and LDA. On the other hand, the signature verification is designed by a novel method based on grid partition, Kernel PCA and LDA. To calculate the similarity between test image and training image, we adopt the selective distance measure determined by various experiments. More specifically, Mahalanobis and Euclidian distance measures are used for face and signature, respectively. As the fusion step, decision rule by weighted sum fusion scheme effectively combines the two matching scores calculated in each biometric system. From the real-time experiments, we convinced that the proposed system makes it possible to improve the security as well as user’s convenience under ubiquitous computing environments.

Prototype classifiers are one of the simplest and most intuitive approaches in pattern classification. However, they need careful positioning of prototypes to capture the distribution of each class region. Classical methods, such as learning vector quantization (LVQ), are sensitive to the initial choice of the number and the locations of the prototypes. To alleviate this problem, a new method is proposed that represents each class region by a set of compact hyperspheres. The number of hyperspheres and their locations are determined by setting up the problem as a set of quadratic optimization problems. Experimental results show that the proposed approach significantly beats LVQ and Restricted Coulomb Energy (RCE) in most performance aspects.

This paper studies fault-tolerance problem of feedforward neural networks implemented in pattern recognition. Based on dynamical system theory, two concepts of pseudo-attractor and its region of attraction are introduced. A method estimating fault tolerance of feedforward neural networks has been developed. This paper also presents definitions of terminologies and detailed derivations of the methodology. Some preliminary results of case studies using the proposed method are shown, the proposed method has provided a framework and an efficient way for direct evaluation of fault-tolerance in feedforward neural networks.

The paper presents a new automated pattern classification method. At first original data points are partitioned by unsupervised self-organizing map network (SOM). Then from the above clustering results, some labelled points nearer to each clustering center are chosen to train supervised generalization regression neural network model (GRNN). Then utilizing the decided GRNN model, we reclassify these original data points and gain new clustering results. At last from new clustering results, we choose some labelled points nearer to new clustering center to train and classify again, and so repeat until clustering center no longer changes. Experimental results for Iris data, Wine data and remote sensing data verify the validity of our method.

The best protein structure prediction results today are achieved by incorporating initial structural prediction using alignments to known protein structures. The performance of these algorithms directly depends on the quality and significance of the alignment results. Support Vector Machines (SVMs) have shown great potential in providing good alignment results in cases where very low similarities to known proteins exist. In this paper we propose the use of a one-class SVM to reduce the computational resources required to perform SVM learning and classification. Experimental results show its efficiency compared to two-class SVM algorithms while producing results of similar accuracy.

Speaker’s intentions can be represented into domain actions (domain-independent speech acts and domain-dependent concept sequences). Therefore, domain action classification is very useful to a dialogue system that should catch user’s intention in order to generate correct reaction. In this paper, we propose a neural network model to determine speech acts and concept sequences at the same time. To avoid biased learning problems, the proposed model uses low-level linguistic features and filters out uninformative features using statistic. In the experiment, the proposed model showed better performances than the previous work in speech act classification. Moreover, the proposed model showed meaningful results when the size of training corpus was small. Based on the experimental results, we believe that the proposed model will be more helpful to dialogue systems because it manages speech act classification and concept sequence classification at the same time. We also believe that the proposed model can alleviate sparse data problems in speech act classification.

In this work, a new phonemes recognition system is proposed. The base of decision of the proposed system is the tongue position and roundedness of the lips. The features of the speech are the coefficients of Wavelet Packet Transform with sub-bands selected through the Mel scale. The SVM (Support Vector Machine) is used as classifier in the structure of a Hierarchical Committee Machine. The database used for the recognition was a set of oral vocalic phonemes of the Portuguese language. The experimental results show success rates of 97.50% for the user-dependent case and 91.01% for the user-independent case. This new proposal increased 3.5% the success rate in relation to the “one vs. all” decision strategy.

In this paper, passport recognition and face verification methods which can automatically recognize passport codes and discriminate forgery passports to improve efficiency and systematic control of immigration management are proposed. Adjusting the slant is very important for recognition of characters and face verification since slanted passport images can bring various unwanted effects to the recognition of individual codes and faces. The angle adjustment can be conducted by using the slant of the straight and horizontal line that connects the center of thickness between left and right parts of the string. Extracting passport codes is done by Sobel operator, horizontal smearing, and 8-neighbornood contour tracking algorithm. The proposed RBF network is applied to the middle layer of RBF network by using the fuzzy logic connection operator and proposing the enhanced fuzzy ART algorithm that dynamically controls the vigilance parameter. After several tests using a forged passport and the passport with slanted images, the proposed method was proven to be effective in recognizing passport codes and verifying facial images.

In this paper, we introduce a modified fuzzy min-max(FMM) neural network model for pattern classification, and present a real-time face detection method using the proposed model. The learning process of the FMM model consists of three sub-processes: hyperbox creation, expansion and contraction processes. During the learning process, the feature distribution and frequency data are utilized to compensate the hyperbox distortion which may be caused by eliminating the overlapping area of hyperboxes in the contraction process. We present a multi-stage face detection method which is composed of two stages: feature extraction stage and classification stage. The feature extraction module employs a convolutional neural network (CNN) with a Gabor transform layer to extract successively larger features in a hierarchical set of layers. The proposed FMM model is used for the pattern classification stage. Moreover, the model is utilized to select effective feature sets for the skin-color filter of the system.