A Study Of Face Recognition Methods Based On Video

Face recognition has long been a hot spot in the field of computer vision.It plays a very important role in the areas of identity authentication, security defense, criminal identification,driving license and passport, the monitoring system of bank and the automatic entrance guard system and so on.General face recognition technology includes two processes of face detection and face recognition.For decades, through the unremitting efforts of researchers, the two processes have been long-term development.Due to the complexity of the application environment of face recognition(such as occlusion, illumination and other factors) and variability of the face itself(e.g. the change of facial expression or aging can cause deformation of face greatly), the research of face recognition has encountered great difficulties, facing new challenges.Aiming at these problems, the paper studied the following two aspects of work, and achieved certain results.Face detection is the first step of recognition, face detection of the traditional detection result is not too ideal, based on skin color detection, increasing the Haar feature extraction,face detection has the accuracy rate is improved; preprocessing is particularly important, in this paper after a face is detected, the detected face region image denoising,normalization preprocessing further, provide a good foundation for face recognition stage.Locally linear embedding(LLE) is a kind of manifold learning method, and has great advantages in dealing with high dimensional image, which remains the main manifold high-dimensional image, eliminate the important parts, greatly improve the velocity of calculation in the recognition process. In this paper, on the basis of LLE, aiming at the shortcomings of ignoring the relativity between samples, the LLE algorithm is proposed based on the Mahalanobis distance, and tested on the Vid TIMIT data set.Experimental results show that the change of calculation method, nearest neighbor using correlation between video frames, can more accurately identify the face, improves the recognition accuracy.