3D face reconstruction from 2D images for face recognition

Face recognition is more preferred to other biometrics, especially in need of identity identification without individuals? direct interaction with a device. Previous studies in this field for over thirty years have shown that non-uniform illumination on the face, the change in facial expression and pose greatly reduce system's success rate. In the literature 3D face recognition systems are proposed for solving such problems. For such recognition systems, 3D face data is needed. 3D face data can be obtained via laser scanners or 2D to 3D reconstruction methods. One of the methods which are recommended for obtaining 3D facial data from 2D facial data is morphable face model. In this study, a different approach is introduced for the process of reconstruction of 3D model using the 2D facial image and the recognition activity is conducted by using model parameters which is obtained through the presented approach. In the proposed method firstly the pose of the face is estimated. At the next step, the environmental conditions and identity parameters which constitute the pose are predicted independently. In this way, the identity parameters will be isolated from the environmental effects which are challenging issues for such systems. Finally, the face recognition process is completed using the identity parameters. The error rates of the previously mentioned steps affect the total success of the system in a cascaded way. Therefore, the success of each step is examined individually. The experiments show that, the novel pose prediction and reconstruction approaches are appropriate for a high performance face recognition system. The proposed method is examined with the CMU-PIE face data set and the success rate of this experiment is 97.9%. For future work, conducting comparison based analysis between 2D views (in different pose and illuminations) generated from 3D model and new 2D face input is planned.