The Study on Face Detection Strategy Based on Deep Learning Mechanism

摘要

In this paper, the deep learning based face detection strategy was proposed. The exploited detection framework has good tolerance to the slight shift of light amplitude and illumination angle. Furthermore, this framework is immune to the slight occlusion. In the detection framework, the CNN was utilized to extract the intrinsic feature and execute the classification on face image and non-face image by recursive convolution and down-sampling process. The convenience of this method lied on that it is not necessary to extract the features explicitly during the detection process. This strategy avoided the information absence due to the inappropriate feature selection. AS a contrast, the LBP-SVM-based feature extraction and classification strategy was utilized to execute the face detection task. The experiment showed the superiority of CNN on detection accuracy and robustness. The comparisons result showed the effectiveness of deep learning mechanism.