Joint Encoding of Multi-scale LBP for Infrared Face Recognition

Due to low resolutions of infrared face image, the local feature extraction is more appreciated for infrared face feature extraction. In the current LBP (local binary pattern) feature extraction on infrared face recognition, single scale is encoded, which consider limited local discriminative information. A new infrared face recognition method based on joint encoding of multi-scale LBP (JEMLBP) is proposed in this paper. To consider correlation in different micro-structures, co-occurrence matrix of multi-scale LBP codes is used to represent the infrared face. The experimental results show the recognition rates of infrared face recognition method based on JEMLBP can reach 91.2% under variable ambient temperatures, outperforms that of the classic method based on single scale LBP histogram.