A Research On Face Recognition Method Based On Geometric Features

This paper introduces the history of the Face Recognition Research as well as the development at home and abroad, and mainly researches on some questions about the automatic face recognition system, including studying the key technologies that has existed, analyzing and summarizing the problems and difficulties in the present studies, and finally focusing on the face recognition methods which are based on the geometric features. To improve on this thesis, the major work I do is summarized as following: In the face image preprocessing part of this paper, I have first introduced some knowledge in Image Enhancement, such as gray-scale transformation, and median filtering. Then I focused on edge detection algorithms, which includes five kinds of the classic algorithms——Roberts, Sobel, Prewitt, LoG and Canny, and three types of improved edge detection algorithms. Comparing the experimental results of these algorithms, I have chosen the improved Canny algorithm at last, and achieved satisfying results. In the automatic face positioning, I have introduced the common integral projection method. And based on this, I adopted a new method which combines the integral projection method and edge detection algorithm together, that is, first to locate the positions of the eyes and mouth in the edge detection binary images, then to get the position of the nose in the gray image using integral projection method. This new method is proved out to be simple and effective. Then I extracted and recorded eleven values featuring the location and size information of the eyes, nose and mouth, and divided them into two feature vectors, one to feature the location, and the other to feature the size. In the last face recognition part, I have adopted an identification method which needs to calculate the Euclidean distance twice——first to match the location vectors, and get the nearest eight images, then to match the size vectors, and output the closest one. To get the necessary human face images, I used part of images in the ORL face database, and the results proved that the identification algorithm is simple in principle, which can reduce the amount of computation, and has achieved fairly good recognition results.