Research On Automatic Generation Of ID Photos And Face Recognition

In recent years,with the development of artificial intelligence technology and the prominent issue of social security,the identity recognition based on human faces has received extensive attention and research.Human face contains a lot of biometric information and the information of faces is often stored in various ID photos.Converting life photos to ID photos by computer vision technology not only facilitates our life,but also generates a large number of face images data which contribute to the research of face recognition.Starting from human faces,the main work of this paper includes:proposing an automatic generation algorithm of ID photos based on prior knowledge to solve the problem that a majority of image segmentation algorithms rely on human-computer interaction.Proposing a face recognition algorithm that integrates the knowledge of face attributes to reduce the influence of illumination,pose and expression changes and to further improve the accuracy of face recognition.The main work and innovation of this paper is as follows:(1)Aiming at the problem that most of image segmentation algorithms rely on human-computer interaction,we propose an automatic generation algorithm of ID photos using prior knowledge.The essence of automatic generation of ID photos is portrait segmentation and the traditional segmentation algorithms rely on human-computer interaction.After analyzing the prior knowledge of the portrait,we propose an automatic segmentation algorithm.The proposed algorithm includes:using face detection to locate the position of the portrait;using edge detection to confirm the outline edge of the portrait;using the built-in template to further confirm the foreground and background areas;using Grab cut algorithm to achieve segmentation results and using matting algorithm to refine the edge.The standard ID photos can be obtained after replacing the background.We verify the effectiveness of the segmentation algorithm on the test set and we also show that there are many other applications of portrait segmentation.(2)Aiming at the problem that there are problems with the existing face recognition algorithms under face expression,posture,and lighting changes,we propose a face recognition algorithm that integrates the knowledge of face attributes.We find that the face attributes such as gender,race,and face shape are robust under face expression,posture,and lighting changes.We use the face attribute recognition model to distill the face attribute knowledge when we train the face recognition model and transfer the face attribute knowledge to the face recognition model by making the mapping of facial feature approach to the output of the face attribute recognition model.The final facial features are both robust and discriminative.We verify the effectiveness of the proposed face recognition algorithm on multiple public datasets and a private dataset under real application scenarios.