Research On 3D Face Reconization Based On Deep Learning

Since nearly half a century, the rapid development of biometric recognition technology has brought great convenience to people’s life. Among them, face recognition with its unique advantages, has been the focus of the majority of researchers. However, face recognition technology in progress at the same time, is also facing many problems to be solved, such as in the process of recognition when using the traditional recognition algorithms, feature selection and extraction becomes the decisive factor to the recognition results; 2D face recognition, lacking of human face’s three-dimensional space information, does not meet the human visual’s characteristic which is 3D and abstract. These problems have become the bottleneck of the development of face recognition technology and the commercial promotion. Due to the limitation of the traditional recognition algorithms and recognition rate tends to be saturated with the increase of the number of samples, scholars turn their attention to deep learning. At the same time, the popularity of cheap RGB-D sensor makes the wide application of 3D face information possible, 3D face recognition research can make full use of the depth information of the face, as well as overcomes the influence of illumination, facial expression, posture, and becomes the research direction in the field of face recognition.Therefore, in this dissertation, the deep learning method is applied to the 3D face recognition technology. A deep learning network is designed to extract features automatically, managing to avoid the impact of feature selection, the depth information of the sample is added on the basis of the original two-dimensional data, and enhance the robustness of the recognition system of illumination, expression and other factors.Firstly, the 3D face point cloud data is transferred to the depth map. Faces are extrated from the 2D images(gray/ color) and depth maps and normalized to improve the efficiency of the system.Secondly, the improved deep convolutional neural networks with soft-max classifier are combined, two features extraction layer networks are designed respectively designed of 2D face image and depth map, database is used to test for the recognition of training and the optimization of the feature extraction is at the principle of the maximization of recognition rate. The output of the two feature layers is used as the input of the recognition layer, and then the training set is used to get the complete Learning Deep recognition network.Finally, the 3D face recognition system which is realized in this dissertation,is tested and compare with other recognition methods to verify the rationality and feasibility of the system. The experimental results show that the recognition method in this paper testing different set of test results with good stability, compared with other recognition methods, the method proposed in this paper on the correct recognition rate has been a substantially increased.