Pretreatment Of Images And Matching Face Recognition

Face recognition has a significant value and wide application foreground in the field of identity authentication. This paper carried out a detailed study and research for preprocessing and matching which was based on the face authentication system and took the second ID cards and scene video images as the research objects.In the preprocessing part, two methods of eyes location were provided firstly, one was taking the gray projection for the gray second ID card image with combination of horizontal and vertical direction, the other was taking template matching for edge image of the normalized scene image and eye images. Then, a method of eyeglasses removal was proposed, the facial image without eyeglasses was synthesized by compensating the occluded region of the facial image with eyeglasses with the corresponding region of the reserved image.In the face matching part, image registration and face matching method are summarized, then, three methods of face matching are discussed in great detail. Firstly, a geometry matching method based on the eyes coordinates finished the matching with coordinate transformation according to calculating the transform parameters which are obtained by eye coordinates of the two facial images. Secondly, a method based on SIFT created the pairs of triangles by using SIFT points which have been matched, then got the registered image with the affine transformation. Thirdly, a template matching method based on statistical feature was finished by extracting the eight directions gradient feature from the fuzzy grid of the image. Finally, the three matching methods were compared using similarity measure in the same standard.In the verification part of preprocessing and face matching, a facial feature library and face images database were built with the second ID cards and scene video images of200people. For the preprocessing effect can be reflected by the effect of the eyeglasses removal method, the effect before and after removing eyeglasses had been compared, the results show that the recognition rate is89.83%and increases by18.82%after eyeglasses being removed. The recognition rate and matching time of the three matching methods had also been compared in the same sample library and experimental environment.