|
计算机应用研究 2012
Log-Gabor and 2D semi-supervised discriminant analysis based face image retrieval
|
Abstract:
Face recognition methods are easily affected by variances of lighting and expression, and there are often only part of sample faces labeled in large image database. To address the two problems, this paper presented a face image retrieval method using multi-channel Log-Gabor wavelet based semi-supervised manifold learning algorithm. Firstly, used Log-Gabor to filter the face image to obtain feature matrix. Then proposed two-dimension semi-supervised manifold learning algorithm to extract the discriminative submanifolds. As the proposed method operated on Log-Gabor feature matrix directly, small sample problem was overcome. Furthermore, by making use of labeled and unlabeled information, preserved the local manifold structure of image data. So enhanced the similarity of feature matching. Experimental results on CMU PIE and AR databases show that the proposed approach is effective and superior to others.