Abstract:
Face recognition has always been a hot topic in the field of computer vision and pattern recognition. With the development of deep learning and the improvement of computer computing performance, a fairly good recognition rate can be obtained on multiple datasets. However, the high recognition rates only exist in the input image sets that are obtained by a standard frontal pose. When face pose changes, the recognition rate will decline rapidly, especially in the case that the face angle is at 45° to 90°. This paper proposes an effective method for pose-invariant face recognition based on convolutional neural network (CNN). This network can project the input face to a high dimensional feature space for explicitly disentangling the identity and pose information in the latent feature space. Different from the conventional deep learning methods that mainly rely on single path CNN structure, the proposed method utilizes the two-pathway CNN structure and metric learning to obtain a feature representation that is invariant to false pose. In the image preprocessing stage, we use the ready-made dlib C++ Library to detect the face and wipe out of the effect of the background. During the model phase, we design a CNN architecture similar to Siamese network for metric learning. Finally, this experiments are made via deep learning framework. The authoritative database is adopted o verify the validity of the proposed model in this paper. Experimental results show that the recognition accuracy of this framework is higher than that of several conventional multi-pose face recognition algorithms.