Adaptive Quantization of Local Directional Responses for Infrared Face Recognition

Abstract

Local feature extraction is one key step in infrared face recognition system. In previous local features extraction (local binary pattern and its variants) on infrared face recognition, a fixed threshold is binary encoded, which consider limited structure information. An infrared face recognition method based on adaptive quantization of local directional responses pattern (AQLDRP) is proposed in this paper. Firstly, the normalized infrared face images are directional filtered to generate local directional responses, which represent the local structures distinctively and are robust to the impacts of imaging conditions. Then, each local feature vector is quantized by adaptive quantization thresholds to preserve distinct information. Finally, the partition histograms concatenation representation is used for final recognition. The experimental results show the recognition rates of proposed infrared face recognition method can reach 93.1 % under variable ambient temperatures, outperform the state-of-the-art methods based on LBP variants.