Improved Sketch-to-Photo Generation Using Filter Aided Generative Adversarial Network

Generating a photographic face image from given input sketch is most challenging task in computer vision. Mainly the sketches drawn by sketch artist used in human identification. Sketch to photo synthesis is very important applications in law enforcement as well as character design, educational training. In recent years Generative Adversarial Network (GAN) shows excellent performance on sketch to photo synthesis problem.  Quality of hand drawn sketches affects the quality generated photo. It might be possible that while handling the hand drawn sketches, accidently by touching the user hand on pencil sketch or similar activities causes noise in given sketch. Likewise different styles like shading, darkness of pencil used by sketch artist may cause unnecessary noise in sketches. In recent year many sketches to photo synthesis methods are proposed, but they are mainly focused on network architecture to get better performance. In this paper we proposed Filter-aided GAN framework to remove such noise while synthesizing photo images from hand drawn sketches. Here we implement and compare different filtering methods with GAN.  Quantitative and qualitative result shows that proposed Filter-aided GAN generate the photo images which are visually pleasant and closer to ground truth image