Researches Of Multi-angle Face Recognition Based On Deep Learning

Multi-angle face recognition has always been a challenge in the field of pattern recognition, which has broad application prospects in the field of military and civilian, and its purpose is to make the computer or other machines automatically recognize multi-angle face. Because multi-angle face is complex and subtle, it will make the face recognition task very difficult and challenging. In recent years, deep learning methods achieved good recognition results in the field of Image Recognition. This paper presents some improved methods and measures after study of the auto-encoder model. The main contents of this paper are as follows:This paper proposes a multi-angle face recognition method based on auto-encoder, which consists of two parts: the first part is the design of face images rotation model based on the auto-encoder model, and the second part is the recognition process of multi-angle face recognition.The design of face images rotation model based on the auto-encoder model consists of two parts: the first part is the initialization of the auto-encoder model. Enter all angles of the face image, from the input layer to the middle layer of the auto-encoder model, and regard the adjacent layers as Restricted Boltzmann Machine, and training each Restricted Boltzmann Machine gradually. Stack all the trained Restricted Boltzmann Machine into the auto-encoder model. Regard weights and bias of each trained Restricted Boltzmann Machine as the weights and bias of the auto-encoder model. The second part is the supervised trimming. Enter all angles of the face image, and tag of the auto-encoder model is a standard frontal face images. Layer by layer using the back-propagation algorithm to adjust the weights and bias of each layer, until the cross entropy of the actual output and the desired output is minimized.The method uses trained auto-encoder model to reconstruct the multi-angle face images into standard frontal face images, and uses these reconstructed standard frontal face images to recognize. The recognition process of multi-angle face recognition consists of two parts. The first part is using these reconstructed standard frontal face images with Euclidean distance, Linear Discriminant Analysis methods for face recognition. The second part is entering the multi-angle face images and the frontal face images into the auto-encoder model, getting the lowest latitude significant features, and using the features for face recognition.At last, various experiments are made in order to test the performance of the auto-encoder model. The results of the experiments show that the auto-encoder model gets good recognition effect, and the identification process can be more intelligent. Some suggestions of the future work are proposed to improve the auto-encoder model.