Research Of Face Recogniton Algorithms Based On Featuere Extracrion

Face recognition technology has been widely studied in pattern recognition, imageprocessing, computer vision and artificial intelligence, etc. Compared with otherbiometric authentication technologies, face recognition technology is characterized bynon-contact, intuitive and tracking, etc. As a result, it has the unique advantages in theapplication.This paper focused on the feature extraction, which is the key problem of thewhole face recognition. The main contents include the following aspects:(1) Feature extraction methods based on the linear space were studied. PCA andLDA feature extraction algorithms were realized on self-built face database. Accordingto the experiment results, deficiencies of the two algorithms were summarized.(2) Since the PCA algorithm with the defects of large amount calculation data,complicated process, as well as poor robustness of light, posture, and facial expression,DCT transform face recognition algorithm based on weighted wavelet was proposed.Face images were decomposed by wavelet, the DCT transform coefficients oflow-frequency and weighted high-frequency were extracted as feature vectors. Theexperimental results show that this method improves the recognition rate and reduce therecognition time.(3) To solve the small sample and edge class projection overlapping problems, anew optimized Fisher standard was adopted. Not only avoiding the requirement ofnonsingular to within class scatter matrix, but also just considering the projectiondirection with redefined weighted between class scatter matrix.