Research On Restoration Algorithm Of Half-tone Security Image Information Based On GAN

Image restoration technology is an important research direction in the field of image processing.With the development of digital technology,it has been widely used in various fields.In order to restore image details,the restoration methods used by researchers are mainly divided into traditional methods and deep learning methods.Traditional image restoration mainly uses the prior features of the image to repair the missing or fuzzy part of the image,which usually faces the unknown prior features,such as the point diffusion function of the imaging system and the power spectrum characteristics of the image before degradation,or complex calculation problems,resulting in the difficulty of image restoration.Deep learning methods use multi-layer networks to simulate the analysis process of human brain neurons,and the restored images and models will learn from a large number of input sample data.However,usually these image restoration technologies focus on solving the macro features of the restored image,and the ultimate goal is to achieve the image satisfying the visual effect of human eyes.However,the dot matrix image is usually less than 30 microns,and its restoration is different from the previous image restoration focusing on macro feature extraction.It is urgent to solve the problem that the information reading of restored image cannot meet the requirements of error detection and correction and decoding.Through the research and analysis of the above problems,this thesis proposes a GAN-based printing micro dots image enhancement and extraction and half-tone security image information restoration,which effectively improves the quality of restored image information and makes the restored image information meet the requirements of error detection and correction and decoding allowed bit error rate.The main research contents of this thesis are as follows:First,the performance evaluation index of the ordinary image restoration method is not enough to evaluate whether the information in the half-tone security image is correctly restored.Therefore,the evaluation index--bit error rate and Euclidean distance,which are suitable for micro image information detection and correction and decoding requirements,is proposed to objectively check the quality of image restoration information.Secondly,in view of problems such as the degradation of image information of printing micro dots in the process of reading secure QR code on mobile phones,the research on image enhancement and extraction algorithm of printing micro dots is proposed.The algorithm is based on GAN to restore the fuzzy image,and then fully utilizes the position features of the printing micro dots in the secure QR code to extract the restored image,realizing the direct generation of printing micro dots image information from the complex carrier image.Thirdly,aiming at the problems of position deviation,insufficient filling and diffusion caused by half-tone security images in the printing process,as well as the quality degradation problems such as strong noise,low quality and distortion caused by different equipment and external environment,an end-to-end restoration model is proposed to take super resolution information of a single image as the target.Firstly,a data set for real printing anti-counterfeiting scenes is constructed to provide reference for the different similarity degree of anti-counterfeiting images encountered in the process of shooting and printing,such as noise and distortion.Then,based on this data set,the high-resolution image information is taken as the restoration target.Firstly,the position tilt of the degraded image is corrected by using the blank line characteristics of the printed quantum dot image,and then the restoration is completed by using ESRGAN’s feature extraction and up-sampling.