Infrared face recognition based on adaptively local directional pattern

Extracting robust and discriminatory features from images is a crucial task for infrared face recognition. For this reason, we have developed an infrared face recognition algorithm based on improved local features, which applies adaptive threshold quantization to encode the local directional energy. The conventional LBP-based feature as represented by the fix threshold encoding has limited distinguishing ability. The adaptive quantization measure of local directional responses can reduce the quantization loss and thus preserve more local structure information in infrared face images. The experimental results under variable ambient temperatures show the recognition rates of proposed infrared face recognition method outperform the state-of-the-art methods based on traditional local features.