The Study On Quality Assessment Algorithm In Face Recognition

With the constantly progress and development of computer technology and digital image processing algorithms of domestic and foreign.Also,after previous development,face recognition was profound changes,which use deal with both face image and video of face stream.It has been widely applied to various fields.For instance security,warning,traffic flow monitoring,authentication and other applications.This paper summarizes the wide-world latest situation that both face recognition technology and influence facial image quality factors.For face image quality assessment algorithm involves key technologies: Lower feature extraction of face image,Classifier.Respectively,it is male a detailed analysis and research.In connection with low and high face image effect on recognition accuracy,Therefore,this paper proposed multilayer neural network classification algorithm based on a deep belief networks,this function is that effectively filter out low-quality facial image identification impact and Improve the recognition accuracy and stability.The main study contents of this paper is fllow: preliminary targeting eye area,eye center coordinate calculation,the offset angle calculating head,face light intensity extraction,face image classification algorithm,with Matlab herein relates to simulate argument.Author Analyzed and summarized recently eye detection algorithm,although Combined with the gray level distribution of the human eye area.Gray eye area detection algorithm based on integral projection was Proposed.By simulation demonstrated,The algorithm is simple and can quickly achieve the detection area.Secondly,the traditional center of the eye location algorithm complexity,computing capacity,and other shortcomings,this paper presents localization algorithm SUSAN filtering algorithm based on the algorithm detected by the human eye to eye two corners farthest positioning Center.Then use the eye center coordinates obtained,according to the spatial positional relationship between the face and the eyes connecting the centers of the normal,the calculated shift angle head.Then,by face illumination compensation algorithm,a low-level extraction algorithm based on image contrast face illumination characteristics were extracted face images are characteristic of the variance and convariance matrix as the light intensity feature.Finally,the traditional shallow supervised classification algorithms cumulative error is too large,the classification accuracy very popular feature of good and bad effects ofdifferent independent feature extraction level,this paper proposes a multi-layer neural network classification algorithm based on a deep belief networks,the algorithm can autonomously transform low-level features to high-level features to achieve a self-extracting features,it has the characteristics of a learning efficiency,and improve the classification accuracy.