Abstract
Several applications including access control, system security need a robust and real-time face identification systems. The last is almost performed by two processes: the detection of the face and then the recognition. In literature, it can be found that some techniques are used exclusively for the detection process and some others are dedicated for the recognition. In this paper, we propose to test and evaluate a common technique for both processes. More particularly, we consider in this paper only the detection process. In order to do that, we have investigated three descriptors: Local Binary Pattern (LBP), Local Binary Pattern Histogram (LBPH), and Histogram Of Gradient (HOG) to get features of faces. These features are used for training process with the same Support Vector Machine classifier taken in different conditions in order to have training dataset. The use of LBPH desciptor followed by SVM classifier for the detection process consitite the second contribution of the paper. Two sets composed of negative images (non-faces) and positive images (faces) are used to generate the training dataset. The test is performed on another dataset. We evaluate the three techniques for face detection on the standard FERET datasets in order to compare the efficiency of these three methods under different face variations and different illuminations. The Local Binary Pattern Histogram with Support Vector Machine classifier outperform the others techniques in terms of efficiency and robustness of face detection. It has been found that the correct detection rate is greater than 98.04% in average.