Key Laboratory of Pattern Recognition and Artificial Intelligence

Laboratory of, China

Key Laboratory of Pattern Recognition and Artificial Intelligence

Laboratory of, China
SEARCH FILTERS
Time filter
Source Type

Huang R.,Sichuan University | Huang R.,Key Laboratory of Pattern Recognition and Artificial Intelligence | Su C.,Sichuan University | Su C.,Key Laboratory of Pattern Recognition and Artificial Intelligence | And 2 more authors.
Journal of Computational Information Systems | Year: 2010

Pseudo-Zernike moments (PZM) feature is widely used in the field of pattern recognition, especially image analysis, due to its lots of good properties such as orthogonality, rotation invariance. However, it is a statistical holistic feature and has the computationally expensive shortcoming for face recognition. In this paper, a new approach, which is called Fisher-weighted PZM (FWPZM) based on Adaboost, is proposed to solve those problems. This approach utilizes a Fisher-weighted function to emphasize the different facial parts. For reduce CPU computation time of FWPZM, Adaboost is used to select the optimal feature set which contains optimal FWPZM orders and corresponding repetitions. FERET face database which contains face images at different orientations, scale, facial expression, different illuminations are selected to evaluate the proposed approach. Experimental results demonstrate the advantages of the proposed method when compared with other PZM face recognition systems. © 2010 Binary Information Press October, 2010.


Huang R.,Sichuan University | Huang R.,South China University of Technology | Huang R.,Key Laboratory of Pattern Recognition and Artificial Intelligence | Su C.,Sichuan University | And 2 more authors.
Journal of Computational Information Systems | Year: 2010

In viewing of the discriminantion feature extraction problem of face recognition, a facial image feature extraction approach based on Gabor wavelet transform and pseudo-Zernike moments is proposed in this paper. This method not only takes into account properties of pseudo-Zernike moment such as invariance, lower information redundant and strongly anti-noise capability, but also considers local space-frequency feature of Gabor wavelet. Firstly the face image convolves Gabor filter bank to obtain Gabor feature. In order to augment the interior relationship of features, they are combined to form Gabor feature image, then orthogonal pseudo-Zernike moment is used to get invariant feature, and linear discriminant analysis is also used to further select features, finally nearest neighbor classification method is adopted to recognize face. Experimental results on the AR face database demonstrate the proposed method has better noise immunity and robust to variations of illumination, pose and facial expression. Copyright © 2010 Binary Information Press.

Loading Key Laboratory of Pattern Recognition and Artificial Intelligence collaborators
Loading Key Laboratory of Pattern Recognition and Artificial Intelligence collaborators