Patch-feature fusion via sparse representation for heterogenous face recognition

Performing a straightforward matching between heterogeneous face images (VIS vs NIR) is really difficult due to high intra-class variability. This paper presents a heterogeneous image transformation(HIT) method based on patch-feature fusion via sparse representation. We first apply photometric preprocessing including Difference of Gaussian filtering to extract local patch structure information. Moreover, gradient information is utilized to enhance the high-frequency content of the synthesized image. By fusing these two kinds of information into the sparse representation model, it is more likely to obtain the optimal reconstruction weight. We will demonstrate our method is efficient for synthesizing high-definition images robust to expression and pose variations. The synthesizing images are competent for later heterogeneous face recognition.