Research On The Method Of Single Sample Face Recognition Based On Feature Fusion In Video Sequences

With the development of artificial intelligence and pattern recognition,face recognition technology has been widely used.Single sample face recognition in video sequence is a typical application of face recognition technology,which recognizes specified face in video sequence based on one single face sample.The recognition accuracy of the existing single sample face recognition method needs to be further improved,which is mainly due to the following two problems.(1)Only one single face sample can be used for each kind of target person,which makes the extraction of facial discriminant feature difficultly.(2)The monitoring scene in the video is not controlled by human constraints,which causes the face to be complicated with the change by different illumination,expression,pose and scale.Address to the above two problems,this paper studies the single sample face recognition from two aspects.This paper focuses on the single sample face recognition of static image and the tracking method of video face and then applies the single sample face recognition of static image to the video to realize the face recognition.The main content of this paper includes following four points:(1)In order to solve the problem of hard to extract discriminant features in single sample face,a single sample face recognition method based on exemplar-svm in static image(FP-Pattern)is proposed.Firstly the whole face is divided into five face sub-patterns with five key points detected by face calibration algorithm.Secondly in the feature extraction of sub-patterns,training classifier of each sub-pattern by exemplar-svm model with external face dataset.(2)Taking into account that each sub-pattern represents the different parts of the face,which have different results to the final face recognition results.An adaptive weighted fusion method of sub-pattern is proposed.This method makes adaptive weighting fusion to the classification result of each person's face sub-pattern and obtains the best recognition object by 1NN classifier.(3)In order to reduce the influence of different pose,expression,illumination and occlusion to the recognition accuracy of the face in video,a single sample face recognition based on feature fusion in video sequence is proposed.The method is divided into two stages: face tracking and face recognition.In the process of face tracking,three single-view face models are merged to obtain a multi-video face model,which is updated incrementally online.In the process of face recognition,according to the different face scale,three face sub-patterns are proposed and combined with FP-Pattern method to get the best recognition object.(4)On the basis of the above research,this paper designs and implements a single sample face recognition system.The system completes the function of face verification(still image),face recognition(video sequence)and so on.The proposed method are compared with existing face recognition methods in five open face datasets: Extend-Yale-B,AR,ORL,Feret-b,ChockPoint.The experimental results show that the proposed method is robust to different illumination,expression,disguises,face angle,scale and has a great improvement in recognition accuracy and efficiency.