Face Sketch-Photo Synthesis and Recognition

This paper presents a simple Face Sketch-Photo Synthesis and Recognition system. Face Sketch Synthesis provides a way to compare and match the faces present in two different modalities (i.e. face-sketches and face-photos). The aim is to significantly reduce the differences between face-sketches and face-photos and also decrease the texture irregularity between them by converting photos to sketches and vice-versa. This results in effective matching between the two thus simplifying the process of facial recognition. This system is modeled using three major components: (i) For a given input face-photo, obtaining an output face-sketch. It is designed using image processing techniques like 2 scale image decomposition and color dodging. (ii) For a given input face-sketch, obtaining an output face-photo. Convolutional Neural Networks are used to model this component. (iii) For a given query face-sketch or face-photo, recognition of face-photo or face-sketch in the database. It is implemented using Fisherface Linear Discriminant Analysis.