Towards Practical Face Recognition: A Local Binary Pattern Non Frontal Faces Filtering Approach

摘要

In dynamic real-time face detection and recognition system, the non frontal faces with different tilt and deflection pose has great influence on the recognition accuracy, in order to solve these problems, we propose non frontal faces filter's method via support vector machine (SVM) and local binary patterns (LBP). By this method the images with large pose deflection will be filtered. Firstly, we apply the AdaBoost algorithm into real-time face detection and join the nose detection to further filter non face images. Then we extract texture feature from the detected face images by LBP feature operator. Finally, SVM is used to classify frontal and non frontal faces. Experimental results show that the proposed method has good classification capability for face images with varying pose. It contribute to eliminate the impact of pose variation in dynamic face recognition system.