Facial Feature Extraction Based On Wavelet Transform And Deformable Template

Facial feature extraction is an important part in the face research field, in this paper extracting facial feature including the outline of the eye and the mouth, is based on the fast wavelet transform and deformable template matching. First of all, the paper reviews the history and current situation, and then analyzes the problems on extracting facial feature using deformable templates; finally, my own research work is discussed. My facial feature extraction includes three important steps: wavelet transform to enhance high frequency, facial feature location and facial feature extraction.Aiming at the problem on extracting facial feature using deformable template, the paper proposes its own development. After locating the face, then do the preprocessing, including mean filter using threshold to get rid of noise and light compensation to reduce the influence of the lighting. Then do wavelet transform using fast wavelet transform algorithm (mallat algorithm), and then facial feature is enhanced by three high frequency parts using different threshold to enhance high frequency.Then we locate feature using template matching, we choose the face images to get eye and mouth templates by myself, using these templates to locate eye and mouth. After location, the eye and mouth are framed by rectangle, we also locate iris centre by analyzing.Finally we extract feature using deformable templates, we design the shape of eye and mouth deformable templates, and the energy functions linking with edge,gradient, corner, weighted variance and average, mouth weighted directed gradient and so on. Parameter ranges of the deformable templates are limited in the rectangle of feature location, and energy function minimum corresponds to the best position of facial feature. We realize the complete facial feature extraction system; the results show the algorithm is feasible to extract the eye and mouth exactly using wavelet transform and deformable templates.