Face Recognition by 2D Wavelet Decomposition and Fisher Discriminant Analysis
Long Fe · 2005
We present a framework of face image representation for appearance--ased face recognition, called SD--WFDA, which is based on 2D Wavelet Decomposition and Fisher Discriminant Analysis (FDA). We first crop face image with the coordinates of two eyes, then decompose it into 2 levels according to Mallat method with a kind of biorthogonal compactly supported wavelet, furthermore, perform Intensity Integral Projection (IIP) to the approximation subband of the second level. Then, FDA, a classificatiorl task specific linear transform technique through supervised learning, is performed to enhance the discriminatory power of extracted features, in which, a Simultaneous Diagonalization (SD) algorithm is adopted to avoid the Possibility of losing discriminative information. The performance of this method is evaluated on YALE and IIS face databases. The experimental results show that the proposed method performs better than traditional Eigenfaces and Fisherfaces approaches. In addition, we expect that this method is also applicable and effective to other Small Sample Size (SSS) pattern recognition tasks.