Cross Domain Descriptor for Face Sketch-Photo Image Recognition

Abstract

Face-sketch to Face-photo matching or recognition is a cross domain modelling problem which identifies the face based on the given sketch query. The sketch and photo differs in their representation so matching the different representations is a challenging task. Compared to hand-crafted image descriptors deep descriptors perform well in solving the problem. Extracting the patches from the sketches and photos can improve the efficiency of the technique. This method is implemented with the benchmarked datasets CUFS, CUFSF and IIITD. The deep descriptor is developed using pre-trained CNN with triplet loss to learn the features. The method performance is better when results are compared with similar procedures.