The authors propose a novel method of fingerprint thinning based on the Gabor-phase image. The Gabor-phase image is an intermediate result of fingerprint enhancing, and the ridge center can be located at the discontinuities of the phase image. The authors utilize an edge-detect-like method to extract the thinned ridges from the phase image. This strategy is different from most existing algorithm which requires extra steps of image binarization so that it can improve both the efficiency and accuracy. Experimental results demonstrate that the proposed algorithm yields better thinning result in much less time.

A novel algorithmusing discriminant analysis of multiple features (DAMF) is presented for adaptive fingerprint image segmentation. Three block-level features includingthe variance，the orientation coherence measure and the total variation are used by the proposed method. For a given fingerprint, probability distribution of each feature over the whole image is estimated, and then a classifier adaptive to the fingerprint is constructed by considering the discriminability of each feature for two-class (background and foreground) classification problem. The fingerprint image is segmented by classifying each image block via the classifier. Mathematical morphology and region boundary smoothingis applied as post processing to obtain compact clusters and to reduce the classification error. The performance of new algorithmis evaluated on FVC2004 database, and the experimental results show that new algorithmcan segment the fingerprints effectively, and also improve the accuracy of matching substantially.

A novel pattern classification method, named AMC-ASVM, was proposed based on Attribute Mean Clustering and Support Vector Machine. The advantages of AMC-ASVM include robustness and high efficiency. AMC ASVM is the natural extension of Cluster Nearest Neighbor Classification as well. Application in Microarray Analysis shows the performance of the present method.

The author proposes a textured image segmentation algorithm which is based on multiresolution wavelet and Gaussian mixture models. At first a preceding process is taken in this algorithm. Then, the image is decomposed by wavelet pyramids. All of those lower resultant image and original image are integrated to a feature vector of a pixel. Those pixels are segmented by Gaussian mixture models. Experiment show the algorithm can give a good result.

Gene selection is a very important problem in microarray data analysis and has critical implications for the discovery of genes related to serious diseases. A fast algorithm based on Fisher Optimization Model (FOM) for gene selection was proposed. The computational complexity of the algorithm relies on the number of the samples not the huge number of genes. Experiments in public data demonstrated the proposed algorithm is effective and efficient.

In this paper a new method for conics based stereo vision was proposed. By using the new method for the detection of general quadric curves proposed by the authors, we investigated the global solutions of planar quadric curves and quadratic surfaces in the space and proposed a unified representation and solving algorithm. The main results about the multi-solutions are as follows: (1) if there exists solution of planar quadric curves then there exists solution of quadratic surfaces; (2) if and only if there are three views then there exists only one solution.