Abstract: We proposed a multitexture centrosymmetric local binary pattern (CSLBP) feature to realize multiview face detection in complex environments. The feature retained the characteristics of Haar ordinal relations, so we drew on the experience of the combination of local binary pattern (LBP) to extract features from four texture directions, such as horizontal, vertical, +45° and -45°, so as to ensure the robustness of face detection in direction, illumination, rotation and so on. The algorithm adopted the cascade architecture. First, face images were partitioned according to different angles of view, and multitexture features were extracted respectively. Then some independent classifiers were designed to eliminate the nonface window step by step. Finally, the multilayer perceptron (MLP) was
used to synthesize the detection effect of each angle of view to output the detection results. The results of verification experiments on data sets FDDB and CMU PIE show that this method is effective for multiview face detection in complex environment. Compared with traditional convolution neural network face detection method, this method has higher accuracy.

Key words: face detection, integration graph, cascade architecture, multitexture centrosymmetric local binary pattern