Human Sketch Recognition using Generative Adversarial Networks and One-Shot Learning
Deepanshu Wadhwa, Utkarsh Maharana, Devina Shah, Vaibhav Yadav, Prashant Pandey · 2019
We introduce a model for face recognition from sketches. The system follows a multi-layered approach and is built by combining sketch to image generation, and face recognition methods. First, sketch to photo generation is achieved by employing cGAN based pix2pix model. Then face recognition is done by One Shot Learning using FaceNet. The sketch recognition model developed through this research is able to yield good results on multiple datasets with the generated images performing with an accuracy close to that with the original images. The average difference from the recognition accuracy as compared to with original images was approximately three percent on the datasets used. Real time recognition using sketches as inputs is also effectively explored through the combination of these two layers of approach.