Catálogo de publicaciones - libros

An approach for tool wear monitoring is presented, which bases on the Feature Vector Selection with Linear Regression (FVS-LR). In this approach, feature vectors are used to capture the geometrical characteristics of tool wear samples, and detection of tool wear is performed by using the model derived from the feature vectors in linear regression method. The signals of cutting force under the condition of tool non-wear and tool wear in 0.6 mm are used to testify the FVS-LR based method for monitoring of tool wear. The results indicate that tool wear can be successfully detected in this method, which is more suitable for the on-line detection in real time because of its efficient algorithm in learning stage and high computing speed in utilization stage.

The performance of human face recognition algorithms is seriously affected by two important factors: head pose and lighting condition. The effective processing of the pose and illumination variations is a vital key for improving the recognition rate. This paper proposes a novel method that can synthesize images with different head poses and lighting conditions by using a modified 3D CANDIDE model, linear vertex interpolation and NURBS curve surface fitting method, as well as a mixed illumination model. A specific Eigenface method is also proposed to perform face recognition based on a pre-estimated head pose method. Experimental results show that the quality of the synthesized images and the recognition performance are good.

Head-and-shoulder detection has been an important research topic in the fields of image processing and computer vision. In this paper, a head-and-shoulder detection algorithm based on wavelet decomposition technique and support vector machine (SVM) is proposed. Wavelet decomposition is used to extract features from real images, and linear SVM and non-linear SVM are trained for detection. Non-head-and-shoulder images can be removed by the linear SVM firstly, and then non-linear SVM detects head-and-shoulder images in detail. Varying head-and-shoulder pose can be detected from frontal and side views, especially from rear view. The experiment results prove that the method proposed is effective and fast to some extent.

Very little research on feature extraction has been taken in the field of network intrusion detection. This paper proposes a novel method of applying principal component neural networks for intrusion feature extraction, and then the extracted features are employed by SVM for classification. The adaptive principal components extraction (APEX) algorithm is adopted for the implementation of PCNN. The MIT’s KDD Cup99 dataset is used to evaluate the proposed method compared to SVM without application of feature extraction technique, which clearly demonstrates that PCNN-based feature extraction method can greatly reduce the dimension of input space without degrading or even boosting the performance of intrusion detection system.

Automatic facial expression recognition has been studied comprehensively recently, but most existent algorithms for this task perform not well in presence of nonlinear information in facial images. For this sake, we employ KPCA to map the original facial data to a lower dimensional space. Then LDA is applied in that space and we derive the most discriminant vectors using GA. This method has no singularity problem, which often arises in the traditional eigen decomposition-based solutions to LDA. Other work of this paper includes proposing a rather simple but effective preprocessing method and using Mahalanobis distance rather than Euclidean distance as the metric of the nearest neighbor classifier. Experiments on the JAFFE database show promising results.

A new ART2A-C algorithm based on fuzzy operators to cluster the remote sensing images and aerials is proposed in this paper. By combining two ART ANNs with higher performance, the traditional ART2A-C is developed with the fuzzy operators introduced in matching rule. Then the proposed method is applied to the classification and the new network is implemented as well as two other existed ARTs respectively. Experimental results show that the new method outperforms the traditional ones.

Color correlogram for content-based image retrieval (CBIR) characterizes not only the color distribution of pixels, but also the spatial correlation of pairs of colors. Color not only reflects the material of surface, but also varies considerably with the change of illumination, the orientation of the surface, and the viewing geometry of the camera. The invariance to these environmental factors is not considered in most of the color features in color based CBIR including the color correlogram. However, pixels changed their color with almost same proportions with change of environmental factors. This fact is taken into consideration, and new algorithm is proposed. The color co-occurrence matrix for different spatial distances is defined based on the maximum/minimum of color component between the three components (R,G,B) of a pixel. The proposed algorithm has less number of features, and the change of illumination, etc. is also taken into account.

A novel data fusion signature verification scheme which combines two schemes is proposed. The first scheme with static features described with Pseudo-Zernike invariant moments and some dynamic features is built with a BP (back-propagation) network. In another scheme, 40 values computed with SVD(singular value decomposition) on thinned signature image and thinned high-density image compose the feature vector and another BP network is built for every kind of signature. Then these two BP networks are connected and their outputs are competitively selected to achieve the final output result. A collection of 290 signatures is used to test the verification system. And experiment shows that FAR (False Acceptance Rate) and FRR (False Rejection Rate) can achieve 5.71% and 6.25% respectively.

The well-known eigenface method uses a single eigenspace to recognize faces. However, it is not enough to represent face images with large variations, such as illumination and pose variations. To overcome this disadvantage, many researchers have introduced multiple eigenspaces into face recognition field. But most of these methods require that both the number of eignspaces and dimensionality of the PCA subspaces are a priori given. In this paper, a novel self-organizing method to build multiple, low-dinensinal eigenspaces from a set of training images is proposed. By in terms of low-dimensional eigenspaces, it completes clustering images systematically and robustly. Then each cluster is used to construct an eigenspace. After all these eigenspaces have been grown, a selection procedure is used to select the ultimate resulting set of eigenspaces as an effective representation of the training images. Then based on these eigenspaces, a framework combined with neural network is used to complete face recognition under variable poses and the experimental result shows that our framework can complete face recognition with high performance.

Quality estimation of fingerprint image can be used to control image quality at the enrollment stage of automatic recognition system and guide the enhancement of fingerprint image. This paper proposes a neural network based fingerprint image quality estimation method. It estimates the correctness of ridge orientation of each local image block using neural network and then computes the global image quality based on the local orientation correctness. The proposed method is used to guide the fingerprint enrollment and improves the accuracy of the automatic fingerprint recognition system.