Catálogo de publicaciones - libros

This research explores the automated detection of surface defects that fall across two different background textures in a light-emitting diode (LED) chip. Water-drop defects, commonly found on chip surface, impair the appearance of LEDs as well as their functionality and security. Automated inspection of a water-drop defect is difficult because the defect has a semi-opaque appearance and a low intensity contrast with the rough exterior of the LED chip. Moreover, the blemish may fall across two different background textures, which further increases the difficulties of defect detection. We first use the one-level Haar wavelet transform to decompose a chip image and extract four wavelet characteristics. Then, the Multi-Layer Perceptron (MLP) neural network with back-propagation (BPN) algorithm is applied to integrate the multiple wavelet characteristics. Finally, the wavelet-based neural network approach judges the existence of water-drop defects. Experimental results show that the proposed method achieves an above 96.8% detection rate and a below 4.8% false alarm rate.

In this paper, a novel graphic symbol recognition of scanned engineering drawing method based on multi-scale autoconvolution transform and radial basis probabilistic neural network (RBPNN) is proposed. Firstly, the recently proposed affine invariant image transform called Multi-Scale Autoconvolution (MSA) is adopted to extract invariant features. Then, the orthogonal least square algorithm (OLSA) is used to train the RBPNN and the recursive OLSA is adopted to optimize the structure of the RBPNN. The experimental result shows that, compared with another affine invariant technique, this new method provides a good basis for the scanned engineering drawing recognition task where the disturbances of graphic symbol can be approximated with spatial affine transformation.

Inspired on psycholinguistics and neuroscience, a symbolic-connectionist hybrid system called - () is proposed, designed to reveal the thematic grid assigned to a sentence. Through a symbolic module, which includes anaphor resolution and relative clause processing, a parsing of the input sentence is performed, generating logical formulae based on events and thematic roles for Portuguese language sentences. Previously, a morphological analysis is carried out. The parsing displays, for grammatical sentences, the existing readings and their thematic grids. In order to disambiguate among possible interpretations, there is a connectionist module, comprising, as input, a featural representation of the words (based on verb/noun classification and on classical semantic microfeature representation), and, as output, the thematic grid assigned to the sentence. - employs biologically inspired training algorithm and architecture, adopting a psycholinguistic view of thematic theory.

The Lernmatrix, which is the first known model of associative memory, is a hetereoassociative memory that presents the problem of incorrect pattern recall, even in the fundamental set, depending on the associations. In this work we propose a new algorithm and the corresponding theoretical support to improve the recalling capacity of the original model.

With the development of Intelligent Transport Systems (ITS), automatic license plate recognition (LPR) plays an important role in numerous applications in reality. In this paper, a coarse to fine algorithm to detect license plates in images and video frames with complex background is proposed. First, the method based on Component Connect (CC) is used to detect the possible license plate regions in the coarse detection. Second, the method based on texture analysis is applied in the fine detection. Finally, a BP Neural Network is adopted as classifier, parts of the features is selected based on statistic diagram to make the network efficient. The average accuracy of detection is 95.3% from the images with different angles and different lighting conditions.

Location of human features, such as human eye centers, is much important for face image analysis and understanding. This paper proposes a context-based method for human eye centers search. A neural network learns the contexts between human eye centers and their environment in images. For some initial positions, the distances between them and the labeled eye centers in horizontal and vertical directions are learned and remembered respectively. Given a new initial position, the system will predict the eye centers’ positions according to the contexts that the neural network learned. Two experiments on human eye centers search showed promising results.

To detect and track moving dim targets against the complex cluttered background in infrared (IR) image sequences is still a difficult issue because the nonstationary structured background clutter usually results in low target detectability and high probability of false alarm. A brand-new adaptive Regularizing Anisotropic Filter based on Partial Differential Equation (RAFPDE) is proposed to detect and track a small target in such strong cluttered background. A regularization operator is employed to adaptively eliminate structured background and simultaneously enhance target signal. The proposed algorithm’s performance is illustrated and compared with a two-dimensional least mean square adaptive filter algorithm and a BP neural network prediction algorithm on real IR image data. Experimental results demonstrate that the proposed novel method is fast and effective.

This study proposes an innovative measure for evaluating the performance of text clustering. In using K-means algorithm and Kohonen Networks for text clustering, the number clusters is fixed initially by configuring it as their parameter, while in using single pass algorithm for text clustering, the number of clusters is not predictable. Using labeled documents, the result of text clustering using K-means algorithm or Kohonen Network is able to be evaluated by setting the number of clusters as the number of the given target categories, mapping each cluster to a target category, and using the evaluation measures of text. But in using single pass algorithm, if the number of clusters is different from the number of target categories, such measures are useless for evaluating the result of text clustering. This study proposes an evaluation measure of text clustering based on intra-cluster similarity and inter-cluster similarity, what is called CI (Clustering Index) in this article.

An object recognition algorithm is put forward based on statistical character of contourlet transform and multi-object wavelet neural network (MWNN). A contourlet-based feature extraction method is proposed, which forms the feature vector taking advantage of the statistical attribution in each sub-band of contourlet transform. And then the extracted features are weighted according to their dispersion degree of data. WNN is used as classifier, which combines the extraction local singularity of wavelet transform and adaptive of artificial neural network. With the application in an aircraft recognition system, the experimental data showed the efficiency of this algorithm for automation target recognition.

Automatic target recpgnition, Wavelet neural network, Contourlet transform, Feature extraction.

In this paper, we propose a face recognition method from a single image per person, called the common subfaces, to solve the “one sample per person” problem. Firstly the single image per person is divided into multiple sub-images, which are regarded as the training samples for feature extraction. Then we propose a novel formulation of common vector analysis from the space isomorphic mapping view for feature extraction. In the procedure of recognition, the common vector of the subfaces from the test face image is derived with the similar procedure to the common vector, which is then compared with the common vector of each class to predict the class label of query face. The experimental results suggest that the proposed common subfaces approach provides a better representation of individual common feature and achieves a higher recognition rate in the face recognition from a single image per person compared with the traditional methods.