Catálogo de publicaciones - libros

In this paper, a simple, yet effective, modification of the activation characteristic and training method of the general fuzzy min-max neural network is presented. The suggested method supplements the hyperbox definition with a frequency factor of the input training patterns. With this factor, a gain value is calculated based on the conception of the relevance of pattern density with respect to the hyperbox range. Utilizing this value, we propose an alternative training method that is able to replace the hyperbox contraction process. Thus, the classification results are independent of the class order presented in the training set. The effect of the gain value is analyzed and the result indicates that the proposed model is also less sensitive to distorted information.

To automatically extract the morphologic features of breast lesion on ultrasound images, K -way normalized cut with the priori rules was applied to perform segmentation, and local area integral invariant was used to analyse the shape of lesion for detecting the structures on the lesion contour and modeling the lesion boundary. Three new morphologic feature measures: mean ( AvgLAI ), standard deviation ( StdLAI ), and signal-to-noise ratio ( SnrLAI ) of the normalized local area integral invariant were proposed to quantify the anfractuosity of lesion shape. Other 132 feature measures were also computed to evaluate the performance of computerized features. These 135 measures characterized the morphologic features, margin features, texture features and acoustic shadowing behavior of the lesions, and were evaluated by ROC analysis on a database of 59 patients. The experimental results showed that the individual morphologic feature measure had the strong ability to distinguish the malignant and benign breast lesions, especially the sensitivity of StdLAI , and SnrLAI could reach 0.92 with the specificity 0.63. It was also found that the discrimination performance of individual feature measure on margin, texture and acoustic shadowing was relatively low on the database.

A novel method is proposed for automatic region-of-interest (ROI) extraction in JPEG2000 coding by using morphology segmentation and analyzing the content of LL subband image. During the JPEG2000 encoding procedure, morphologic filter group is implemented on the wavelet LL subband image and the watershed algorithm is applied to divide it into many closed regions rapidly following the discrete wavelet transform. Afterward, certain local texture features of regions are extracted to merge the little regions and then the regions location features are employed to determine the ROI of the LL image and figure out the LL ROI template. Finally the entire ROI template of wavelet image is formed by the LL ROI template according to the discrete wavelet transform (DWT) structure and the coordinate mapping principle. This method realizes the automatic JPEG2000 ROI coding based on the content of image. The experiment results have proven its feasibility and validity.

Chinese characters cognition is an important field of cognitive science. So more attention has been paid to the research of Chinese characters cognition. And an effective method is computer simulation. In this paper, a model that simulates the clustering of Chinese characters and the splitting of components based on self-organizing map is proposed. In other words, it can research the relation of Chinese characters and the components. This is the important research content of Chinese characters cognition. Results from this simulation suggest that the model is able to account for some empirical results.

The automatic design algorithm of the bottleneck is the critical problem to realize computer-aided design of railway yard. Each station bottleneck can be decomposed two components: a core switches group and several ladder tracks that are composed by six basic equipments. The structure of the sketch of a bottleneck can be input and then recognized by means of optimized mathematic model. And all the arrangement style of the ladder track can be enumerated by means of different morphology of a binary tree. The combination method of the sketch input and structural enumeration developed in the paper makes it available to design railway yard in an automatic and optimized method based on constrains reasoning optimum mathematical models.

Iris recognition is a kind of important biometrics technology for personal identify verification, iris classification method has been achieved more attention according to different feature extraction. Binary feature extraction is one of the most effective techniques employed for the human iris recognition problem. However, the selection of a particular set of features is often problematic, so iris classifier performance isn’t satisfied. Based on AdaBoost, an enhanced algorithm for iris classifier is presented in this paper. The algorithm will achieve a stronger iris classifier (iris feature template) by lifting weaker similarity classifiers based on AdaBoost through training samples. Simulation results on CASIA iris database show that the method is effective.

The problem considered in this paper is the automatic microscopic image mosaicing. In this paper, invariant local features based on normalized moment of inertia are used to select matching points and calculate the translation. This algorithm consists of feature detection, feature registration and spatial transition. The experimental results with more than 100 clinical medical images in different categories demonstrate this algorithm is fast, effective and does not require human interaction.

This paper presents an iterative method for tracking and classifying human activities in a video sequence. The basic idea is that activities can be positively identified from a sparsely sampled sequence of a few body poses acquired from videos. Connected Graph representation has been used to store the 2D human poses. These samples are matched against the graph abstractions derived from the frame where motion is identified in the video sequence. Sum of Absolute Differences method is used for motion detection in video frames. The probability of false activity detection drops exponentially with the increased number of sampled body poses. The proposed method gives very good results for activity detection in the surveillance video.

Ringing artifacts arise near large edges in highly compressed JPEG2000 images. In this letter, a ringing artifact reduction scheme designed for low bit-rate JPEG2000 images is proposed. A local adaptive filter is adopted to process the received image in spatial domain and each wavelet subband. Meanwhile, a new method to estimate the variance of compression loss is put forward. The scheme can achieve both PSNR improvement and human perceptual enhancement in most cases.

We propose a shape recognition method for the fast retrieval of objects in 2D images. The algorithm is based on recent developments in support vector machines and skeleton match. The shape recognition method is robust in the presence of noise and is irrespective of variations in rotation, scale and translation. The method has been implemented and performed experiments on some image data. Our experimental results showed characteristics of our method. In the end, the future research directions are discussed.