Feature and keypoint selection for visible to near-infrared face matching

Matching near-infrared to visible images is one of the heterogeneous face recognition challenges in which spectral variations cause changes in the appearance of face images. In this paper, we propose to utilize a keypoint selection approach in the recognition pipeline. The proposed keypoint selection approach is a fast approximation of feature selection approach, yielding two orders of magnitude improvement in computational time while maintaining the recognition performance with respect to feature selection. The keypoint selection approach also enables to visualize the keypoints that are important for recognition. The proposed matching framework yields state-of-the-art approaches results on CASIA NIR-VIS-2.0 dataset.