Catálogo de publicaciones - libros

Automatic reconstruction of patient-specific 3D bone model from a limited number of calibrated X-ray images is not a trivial task. Previous published works require either knowledge about anatomical landmarks, which are normally obtained by interactive reconstruction, or a supervised initialization. In this paper, we present an automatic 2D/3D reconstruction scheme and show its applications to reconstruct a surface model of the proximal femur from a limited number of calibrated X-ray images. In our scheme, the geometrical parameters of the proximal femur are obtained by using a Bayesian filter based inference algorithm to fit a parameterized multiple-component geometrical model to the input images. The estimated geometrical parameters are then used to initialize a point distribution model based 2D/3D reconstruction scheme for an accurate reconstruction of a surface model of the proximal femur. Here we report the quantitative and qualitative evaluation results on 10 dry cadaveric bones. Compared to the manual initialization, the automated initialization results in a little bit less accurate reconstruction but has the advantages of elimination of user interactions.

Based on neural network with favorable adaptability to handwritten Chinese character multi-features, in this paper a new method is proposed, using existing multi-features as inputs to structure multi neural network recognition subsystems and these subsystems are integrated with parallel connection mode. The integrated system has the lowest false recognition rate. When using traditional von Neumann architecture computer to implement this system, the system response time is longer as a result of serial computation. This paper introduces a kind of parallel computation method of using pc cluster to implement multi subsystems. It can reduce effectively recognition system’s response time.

This paper proposes an algorithm based on fuzzy support vector machine (FSVM), a new pattern analysis method, for detecting the abnormal trajectory patterns of moving objects from surveillance video. Firstly, feature points are extracted for presenting continuous trajectories. Then fuzzy memberships are introduced to measure contributions of the feature points of trajectory. Finally, the algorithm is applied to detect the abnormal patterns in 2D object trajectories. Experiments on trajectory data set show the validity of the algorithm.

In speaker recognition fields, score normalization is a widely used and effective technique to enhance the recognition performances and is developing further. In this paper, we are focused on the comparison among many kinds of candidates of score normalization methods and a new implementation of the speaker adaptive test normalization (ATnorm) based on a cross similarity measurement is presented which doesn’t need an extra corpus for speaker adaptive impostor cohort selection. The use of ATnorm for the language robustness of the multi-language speaker verification is also investigated. Experiments are conducted on the core task of the 2006 NIST Speaker Recognition Evaluation (SRE) corpus. The experimental results indicate that all the score normalization methods mentioned can improve the recognition performances and ATnorm behaves best. Moreover, ATnorm can further contribute to the performance as a means of language robustness.

In this paper, a novel face recognition method based on binary face edges is presented to deal with the illumination problem. The Binary Face Edge Map (BFEM) is extracted using the Locally Adaptive Threshold (LAT) algorithm. Based on BEFM, a new image similarity metric is proposed. Experimental results show that face recognition rates of 76.32% and 82.67% are achieved respectively on 798 AR images and 150 Yale images with changed lighting conditions and facial expression variations when one sample per subject is used as the target image. The proposed method takes less time for image matching and outperforms some existing face recognition approaches, especially in changed lighting conditions.

This work introduces a new approach to detect fake fingers, based on the analysis of time-series fingerprint images. When a user puts a finger on the scanner surface, a time-series sequence of fingerprint images is captured. Five features are extracted from the image sequence. Two features represent the skin elasticity, and three features represent the physiological process of perspiration. Finally the Support Vector Matching (SVM) is used to discriminate the finger skin from other materials such as gelatin. The experiments carried out on a dataset of real and fake fingers show that the proposed approach and features are effective in fake finger detection.

In this paper, the line segment tracking is considered for an imaging system with translational motions (T.M.), rotational motions (R.M.) and arbitrary motions. Assuming a CCD camera with an arbitrary tilt angle installed on a mobile robot equipped with odometry system, to reduce the search space in the correspondence problem, two constraints are developed. These constraints are location and orientation differences (O.D.) for line segments in consecutive images. The findings of this paper include: 1)The development of the effect of camera tilt on the location constraint for T.M., 2)Illustrating that the upper bound of O.D. for both horizontal and vertical lines with respect to the floor, is a function of tilt in T.M., which can considerably be reduced and thus providing a tight constraint, 3)The development of the location constraint for R.M., 4) The development of the PDF and upper bound of O.D. for R.M., 5) The development of the location and O.D. constraint for arbitrary motions. Furthermore, the efficiency of these developed constraints in a line tracking algorithm was verified.

Content-based image classification has always been a hot research topic. This paper aims to propose an efficient image analysis algorithm using geometric features of skin regions to effectively classify images. First, a nonparametric skin color classifier is used to skin detection. Then, the contours of skin regions are constructed using a curve evolution method based on adaptive grids. Finally, the geometric features are extracted from the contours, and the cosine similarity measure is adopted for image classification. The algorithm is tested on a large database consisting of 6000 images. Experimental results illustrate the proposed method perform well in classifying skin images.

In many cases human identification biometrics systems are motivated by real-life criminal and forensic applications. Some methods, such as fingerprinting and face recognition, proved to be very efficient in computer vision based human recognition systems. In this paper we focus on novel methods of human identification motivated by the forensic and criminal practice. Our goal is to develop computer vision systems that would be used to identify humans on the basis of their lips, palm and ear images.

The quality of character image plays an important role for the performance of character recognition system. However there is no good way to measure the recognition difficulty of a given character image. For the given character image with unknown quality, it is improper that apply the single character database to recognize it by the same feature and the same classifier. This paper proposed a novel approach for multi-resolution character recognition whose feature is extracted directly from gray-scale image and classification is adaptive classification which adaptively selects the appropriate character database and classifiers by evaluating the image quality of the input character. A resolution evaluation algorithm based on gray distribution feature was proposed to decide the adaptive classification weights for the classifiers, which make the classification have the higher probability of being the correct decision. Experiment results demonstrate the proposed approach highly improved the performance of character recognition system.