Research On Restoration Technology Of Simulated Sketch Portraits Based On Deep Generative Adversarial Networks

Simulated sketch portraits have significant application in domestic and international crime detection.Portrait artists draw sketch portraits of criminal suspects based on the witness’ s description of face,hair,posture and other portrait features,which help narrow the scope of the suspects or determine the suspects through mass identification.Simulated sketch portraits are subjectively created by portrait artists.The thickness and density of lines are used to express face information.The lack of color,feature details and other information leads to large difference between portraits and real face photos.Deformation,gray-scale differences and lack of textures have brought more difficulties to the actual identification work.In order to improve the applicability of simulated portraits,this thesis proposes a restoration technology based on a generation adversarial network,which restores sketched portraits into high-resolution color facial images,and controls the facial features of the generated images.The contents of this thesis are as follows:First of all,this thesis studies the sketch portrait recoloring methods including traditional image pseudo-color processing technology and the image style transfer algorithm based on the generation adversarial networks,and optimize the BicycleGAN algorithm according to the characteristics of sketch simulation portrait conversion task and to convert the sketch portrait into a color image close to a real human face.Secondly,this thesis studies the current mainstream face image editing algorithm,optimizes the STGAN algorithm through orthogonalization of feature vectors to remove the correlation between feature vectors to realize face feature editing of single feature.Subsequently,this thesis studies the current image super-resolution reconstruction algorithm,compares a variety of face image resolution improvement algorithms through intuitive judgment and image quality evaluation indicators,and uses the ESRGAN algorithm to perform super-resolution reconstruction of face images.Finally,this paper builds a software for color reproduction and feature adjustment of sketch portraits for practical use.The research results show that the sketch portrait restoration technology based on generation adversarial network can restore sketch portraits into photo realistic faces,and can flexibly adjust the facial features in restored facial images.