Catálogo de publicaciones - libros

In order to realize robust visual tracking in natural environments, a novel algorithm based on adaptive appearance model is proposed. The model can adapt to changes in object appearance over time. A mixture of three Gaussian distributions models the value of each pixel. An online Expectation Maximization (EM) algorithm is developed to update the parameters of the Gaussians. The observation model in the particle filter is designed based on the adaptive appearance model. Numerous experimental results demonstrate that our proposed algorithm can track objects well under illumination change, large pose variation, and partial or full occlusion.

The paper presents and realizes a new image retrieval method based on the combination of the color connected area information with the texture features. The image is firstly divided into several parts and the color connected areas in the image is computed, then, the primitive co-occurrence matrix of the four color components corresponded with the connected area of each color is extracted. Lastly the image retrieval on the basis of the content is realized by using of the feature similar function which is designed according to these features.

Fault diagnosis is a small sample problem as fault data are absent in the real production process. To tackle it, Support Vector Machines (SVM) is adopted to diagnose the chemical process steady faults in this paper. Considering the high data dimensionality in the large-scaled chemical industry seriously spoil classification capability of SVM, a modified adaptive chaotic binary ant system (ACBAS) is proposed and combined with SVM for fault feature selection to remove the irrelevant variables and ensure SVM classifying correctly. Simulation results and comparisons of Tennessee Eastman Process show the developed ACBAS can find the essential fault feature variables effectively and exactly, and the SVM fault diagnosis method combined with ACBAS-based feature selection greatly improve the diagnosing performance as unnecessary variables are eliminated properly.

In this paper, we present a novel eye location approach based on image context analysis. It is robust from the image variations such as illumination, glasses frame, and eyebrows. Image context of an image is any observable relevant attributes with other images. Image context analysis is carried out using the hybrid network of k-means and RBF. The proposed eye location employs context-driven adaptive Bayesian framework to relive the effect due to uneven face images. The appearance of eye patterns is represented by Haar wavelet. It also employs a merging and arbitration strategy in order to manage the variations in illumination and geometrical characteristics of ambient eye regions due to glasses frames, eye brows, and so on. The located eye candidates are merged or eliminated, and adaptive arbitration strategy is used based on a minimizing energy function by probabilistic forces and image forces. The adaptation is carried out by the analysis of image context. The experimental results show that the proposed approach can achieve superior performance using various data sets to previously proposed methods.

This paper proposes a robust hand-posture recognition method by learning similarity between hand-posture and structure for the performance improvement of vision-based hand-posture recognition. The difficulties in vision-based hand-posture recognition lie in viewing direction dependency and self-occlusion problem due to the high degree-of-freedom of human hand. General approaches to deal with these problems include multiple camera approach and methods of limiting the relative angle between cameras and the user’s hand. In the case of using multiple cameras, however, fusion techniques to induce the final decision should be considered. Limiting the angle of user’s hand restricts the user’s freedom. The proposed method combines angular features and appearance features to describe hand-postures by a two-layered data structure and includes learning the similarity between the two types of features. The validity of the proposed method is evaluated by applying it to the hand-posture recognition system using three cameras.

A novel scheme is proposed for training Support Vector Machines (SVMs) in automatic recognition of young-old gait types with a higher accuracy. Kernel-based Principal Component Analysis (KPCA) is employed to initiate the training set, which efficiently extracts more nonlinear features from highly correlated time-dependent gait variables and improves the generalization performance of SVM. With the proposed method (abbreviated K-SVM), the gait patterns of 24 young and 24 elderly normal participants were analyzed. Cross-validation test results show that the generalization performance of K-SVM was on average 89.6% to identify young and elderly gait patterns, compared with that of PCA-based SVM 83.3%, SVM 81.3% and a neural network 75.0%. These results suggest that K-SVM can be applied as an efficient gait classifier for young and elderly gait patterns.

In this paper, an interactive color planning system is proposed. The core of the system is an emotion-based pattern retrieval that can be implemented with two data bases. In the system a user can issue an adjective word that represents the desired feeling to the system. Then the system retrieves the color combinations or representative color patterns from the knowledge base. Again when the user selects a color combination or representative color pattern, the system provides similar product patterns from the product DB according to MPEG-7 visual descriptors. Finally the user can choose the color patterns and apply them to the 3D VRML space. The decision support system can help users who do not have enough knowledge about color planning and can be used to promote e-business.

A novel algorithm applied linear program to estimate the generalization performance of SVM is presented. When span is used to estimate the generalization performance of SVM, a series of quadratic programs needs to be solved, of which the object function defines an elliptic norm. Based on the theorem of convergence property of norm, the function can be approximated to an infinity norm, and then a linear program is achieved. The theoretic analysis and experiment results show that the method can estimate the generalization performance well and reduce the computation time greatly.

A good size measurement method should meet at least three criteria. These are rotational invariance, reproducibility and embody overall shape description (elongation / flakiness or angularity). According to these three criteria, this paper analyzes and evaluates several existing methods of solid particle measurement, such as Chord sizing, (multiple) Ferret diameter, equivalent circle, maximum diameter and equivalent ellipse etc. in image analysis. Based on the analyses and evaluations of the existing methods, we propose a new method – best-fit rectangle for size measurement that satisfactorily meets the criteria of rotational invariance, reproducibility and shape description.

Nowadays, empirical mode decomposition (EMD) and Hilbert spectrum (HS) have been broadly investigated on non-stationary and nonlinear signal processing, especially on vibration signal analysis. But as diesel engine vibration signal wide frequency band, it leads to this method not decompose intrinsic mode function (IMF) successfully. Therefore, the obtained IMF is less meaning. For a better recognition of diesel condition because of its wider frequency band, this paper uses wavelet packet as preprocessing for HS analysis. It can effectively reduce wide frequency band and noise interference. Thus, the developed method is named as Wavelet Packet Hilbert Spectrum (WPHS). Experimental data of a DI135 diesel engine with different fuel supply advanced angle is used to evaluate effectiveness of the developed methodology on diesel pattern recognition. According to the recognition result, it can be concluded that this approach is very promising for reciprocating engine condition classification and preventative maintenance.