一种具有较好光照鲁棒性的人脸识别方法

In the face recognition system, the illumination condition is a very important issue. In order to effectively overcome the influence of illumination change, this paper, an improved local binary pattern method is proposed. Firstly, neighborhood mean is used as a threshold to obtain better description of the local texture. Secondly, in order to better describe and reflect the large and complex parts of a face image, meanwhile reduce some redundant patterns, we use 99% of all occurred patterns of the LBP patterns as the final patterns. The original image is processed based on the final patterns, the ILBP image is obtained. Finally we use principal component analysis (PCA) for classification. Experiments based on AR and Yale face database show that the ILBP image combined with PCA can significantly improve recognition accuracy, especially has strong robustness to illumination.