A novel technique for sketch to photo synthesis

Abstract

We propose a novel pseudo photo generation system from a sketch image. All the training images and the input test sketch are warped around its mean shape to get shape free images. Based on local geometry preserving manifold learning method called locally linear embedding (LLE), a shape free pseudo-photo is synthesized by block matching. Then the photo of actual shape is constructed with the help of an Active Shape Model(ASM) and a neural network built on shape control point coordinate pair for training sketch-photo samples. We experimented over 300 sketch-photo image pairs and output synthesized photo for every test image is encouraging. � 2010 ACM.