Optimization Of Face Image Sequence Based On MTCNN

In this thesis,the research results of the frontier neural network algorithm,combined with the tracking algorithm,on this basis,combined with skin color detection,clarity detection,face pose estimation,build a face detection,tracking,optimization for image sequence as a whole,I believe it can be used in human face recognition system and has a good improvement effect.Aiming at the problem of extracting the face region in the image,the MTCNN network algorithm is used as the face detection algorithm,which can effectively and quickly detect the face part in the image.In the image sequence,the position of the face may change continuously with time.Aiming at the problem that the face area cannot be dynamically locked,the improved Camshift algorithm is used for face tracking,which can also be used in the process of face movement.Can firmly lock the head area.In response to the needs of most face recognition algorithms for the image to be recognized,in order to achieve the best and best facial image for its recognition,so as to improve the efficiency of face recognition,skin color detection is added after the face image sequence is obtained for occlusion.Screening,the image definition algorithm performs three steps of image definition screening and face pose estimation screening frontal images to ensure that the image sequence obtained by the face recognition module is easy to detect and recognize,avoid wasting unnecessary computing power,and improve the work of recognition applications effectiveness.