The Research Of Face Recognition System Based On Embedded Platform

With the development of embedded technology and face recognition, face recognition is widely used in real life, however, most embedded face recognition system is used commercially, its cost is relatively high. In this paper, the face detection, preprocessing, face feature extraction and recognition algorithm process and the efficiency was studied, then select the algorithm which is very good on the embedded platform and improve the efficiency of the algorithm. The algorithm experiments are conducted on Windows, Linux platform, then generate the algorithm which can be used in multiple platform. And the algorithm was applied to the Android platform, to develop a practical face recognition system, realizing the efficient and accurate face recognition under small sample, and tests in Android4.2.1 versions of HuaWei G610 cell phones have achieved good effect.This paper studied the face detection algorithm,and puts forward a double eye detection, and do the eyes accurate detection in the eyes rough areas. Actual tests, up and down around the face can be quickly and accurately detected. In order to solve face angle rotation problem, face images acquisition is based on the multiple perspectives of face, on the left side of the face, the right side of the face, the up face and the down face.In the local feature extraction, face feature pool is established on various angles faces, when face recognition feature points will match with feature pool features.In ordinary life, to identify a person look at the whole contour is similar, then looking closer local features is similar. According to the habits, the triple face recognition is proposed in this paper, the first step is to get the face’s global characteristics of gradient (HOG feature), then using the SVM classifier for training and classification. The overall gradient characteristics can be used to predict the faces of the match of unknown sample library; The second face recognition uses the improved SURF algorithm to collect the local characteristics, then uses k neighbor classifier to get the face’s classification. This paper puts forward regional characteristics acquisition and regional feature matching which can improve the matching accuracy and the matching efficiency; The local feature matching often encounter the problem of setting the threshold, too strict will reduce the recognition rate, too loose will result in an increase in mistakenly identified. Aiming at this problem, this paper puts forward the third face recognition, when the second recognition result is around the threshold value. On the third face recognition, extracting the local characteristics of T area, then generates the local binary pattern texture descriptor based on Gabor feature, eventually, uses k neighbor classifier to get the classification. This paper introduced the triple face recognition judgment, but efficiency is higher, the first step to identify requires the average time of 0.023 seconds, the second just compared with prediction of the user, time is short, the average time of 0.298 seconds, the third based on T area detection, speed faster, the average time of 0.009 seconds. Face recognition algorithm takes about 0.6 seconds on average, the overall recognition time within 1 s, satisfies the requirement of real-time performance.This paper designs and realizes a face recognition system, the system combines facial recognition technology and user information management, provide the administrator and ordinary users module, administrator can manage users, see a stranger, to obtain a stranger or user login SMS, etc.; After facial recognition, ordinary users can manage personal information and manage personal photo albums, etc.