Catálogo de publicaciones - libros

We consider the security of the Rila-Mitchell security protocols recently proposed for biometrics-based smartcard systems. We first present a man-in-the-middle (MITM) attack on one of these protocols and hence show that it fails to achieve mutual authentication between the smartcard and smartcard reader. In particular, a hostile smartcard can trick the reader into believing that it is a legitimate card and vice versa. We also discuss security cautions that if not handled carefully would lead to attacks. We further suggest countermeasures to strengthen the protocols against our attacks, as well as to guard against the cautions highlighted. Our emphasis here is that seemingly secure protocols when implemented with poor choices of parameters would lead to attacks.

The motor is the workhorse of industries. The issues of preventive and condition-based maintenance, online monitoring, system fault detection, diagnosis, and prognosis are of increasing importance. This paper introduces a technique to detect faults in induction motors. Stator currents are measured by current meters and stored by time domain. The time domain is not suitable for representing current signals, so the frequency domain is used to display the signals. Fourier transform is used to convert the signals onto frequency domain. After the signals have been converted, the features of the signals are extracted by the signal processing methods like the wavelet analysis, spectrum analysis, and other methods. The discovered features are entered to a pattern classification model such as a neural network model, a polynomial neural network, a fuzzy inference model, or other models. This paper describes the results of detecting fault using Fourier and wavelet analysis.

Geodesic active contour is a useful image segmentation method. But it may fail to segment objects disturbed by complex noises. Prior knowledge on certain object is a powerful guidance in image segmentation. We represent region and shape prior of certain object in a form of speed field and incorporate it into Geodesic Active Contours. Region prior constrains the zero level set evolving in certain region and shape prior pulls the curve to the ideal contour. Applications in a large quantity of cardiac valve echocardiographic sequences have shown that the algorithm is a more accurate and efficient image segmentation method.

The canonical form of many large-scale scientific and technical computing problems are often linear algebra problems. As such, routines such as matrix solvers find their use in a wide range of applications. The performance of matrix solvers are often critical in determining the performance of the application programs. This paper investigates the performance of common linear algebra routines on the current architectures of interest to supercomputing users, namely the Intel Xeon EM64T and ItaniumII, with examples from OptimaNumerics Libraries. Performance issues and myths are also shown and diffused in this paper.

Model selection plays a key role in the performance of support vector machines (SVMs). At present, nearly all researches are based on binary classification and focus on how to estimate the generalization performance of SVMs effectively and efficiently. For problems with more than two classes, where a classifier is typically constructed by combining several binary SVMs [8], most researchers simply select all binary SVM models simultaneously in one hyper-parameter space. Though this method works well, there is another choice – the method where each binary SVM model is selected independently and separately. In this paper, we compare the two methods for multi-class SVMs with the strategy [8]. Their properties are discussed and their performance is analyzed based on experimental results.

A License plate recognition (LPR) system can be divided into the following steps: preprocessing, plate region extraction, plate region thresholding, character segmentation, character recognition and post-processing. For step 2, a combination of color and shape information of plate is used and a satisfactory extraction result is achieved. For step 3, first channel is selected, then threshold is computed and finally the region is thresholded. For step 4, the character is segmented along vertical, horizontal direction and some tentative optimizations are applied. For step 5, minimum Euclidean distance based template matching is used. And for those confusing characters such as ’8’ & ’B’ and ’0’ & ’D’, a special processing is necessary. And for the final step, validity is checked by machine and manual. The experiment performed by program based on aforementioned algorithms indicates that our LPR system based on color image processing is quite quick and accurate.

We present a parallel approach for optimizing surface meshes by redistributing vertices on a feature-aware higher-order reconstruction of a triangulated surface. Our method is based on a novel extension of the fundamental quadric, called the . This quadric helps solve some basic geometric problems, including detection of ridges and corners, computation of one-sided normals along ridges, and construction of higher-order approximations of triangulated surfaces. Our new techniques are easy to parallelize and hence are particularly beneficial for large-scale applications.

Mobile data mining [8-11] is about the analysis of data generated by mobile activities, in search for useful patterns in order to support different types of decision making requirement. Mobile devices are loaded with features such as the capability to listen to radio from a mobile phone. Mobile users who listen to radios on their mobile phones are a source of data generated from mobile activities. The location dependent data [9] and the song they listen to can be combined and analysed in order to better understand the behaviour of mobile users. This paper shows how this can be done by using taste template, which categorises a behaivoural type in order to match mobile users into one of these categories. Conclusion from this research project confirms a new way to learning behaviour of mobile users.

A view validation algorithm has been shown to predict whether or not the views are sufficiently compatible for solving a particular learning task. But it only works when a natural split of features exists. If the split does not exist, it will fail to manufacture a feature split to build the best views. In this paper, we present a general algorithm (Correlation and Compatibility based Feature Partitioner) to automate multi-view detection. CCFP first labels the large amount of unlabeled examples using single view algorithm, then calculates the conditional SU (Symmetric Uncertainty) between every pair of features and the IG (Information Gain) of each feature given the examples labeled previously by single view algorithm with high-confidence predictions. According to the estimated values of SU and IG, all the features will be partitioned into two that are low correlated, compatible and sufficient enough. The experiment results show that multi-view learner with views generated by CCFP outperforms learner with views generated by other means clearly.

We present a neuro-dominance rule for single machine total weighted tardiness problem. To obtain the neuro-dominance rule (NDR), backpropagation artificial neural network (BPANN) has been trained using 5000 data and also tested using 5000 another data. The proposed neuro-dominance rule provides a sufficient condition for local optimality. It has been proved that if any sequence violates the neuro-dominance rule then violating jobs are switched according to the total weighted tardiness criterion. The proposed neuro-dominance rule is compared to a number of competing heuristics and meta heuristics for a set of randomly generated problems. Our computational results indicate that the neuro-dominance rule dominates the heuristics and meta heuristics in all runs. Therefore, the neuro-dominance rule can improve the upper and lower bounding schemes.