Face Photo Recognition from Sketch Images Using HOG descriptors

Abstract

In this paper, we propose an efficient algorithm for matching sketches with face-photo images. The proposed system extracts face photo from the database based on a query sketch drawn by an artist by using discriminating information present in the facial regions represented as histogram of oriented gradients (HOG) feature descriptor. From the training set consisting of photo-faces, HOGs features are extracted and stored as knowledge base. Given, an artist sketch, HOGs of the sketch are computed and are compared against the knowledge base by applying KNN classifier for retrieval of best matching photo face. In our study we have considered face photos of frontal pose with normal lighting and neutral expression. No occlusions in the photo are assumed. Experiments are conducted on CUHK student database of face sketches and photos.