Research On Reconstruction Method Of Incomplete Fingerprint Image Based On Adversarial Generative Network

Fingerprint recognition technology is one of the most important technologies in the field of biometrics and has been widely used.At present,the recognition of high-quality fingerprint images is very mature,but there is still a bottleneck in the recognition of incomplete fingerprint images.There are few researches on incomplete fingerprint image reconstruction technology in my country.Traditional image algorithms are mainly used to reconstruct incomplete fingerprint images in small areas.However,for fingerprint images with a large degree of incompleteness,traditional reconstruction methods cannot obtain better reconstruction results.Therefore,it is of great significance to study the reconstruction method of incomplete fingerprint images.This thesis aims to use the adversarial generative network to reconstruct the incomplete fingerprint images.The main contents are as follows:(1)Aiming at the problem that traditional fingerprint image reconstruction methods cannot repair large-area incomplete fingerprint images,this thesis improves a residual-based Shift-Net network model for incomplete fingerprint image reconstruction.Three consecutive residual blocks are added to the Shift layer of the original Shift-Net network,which migrates the position of the Shift layer in the original Shift-Net and increases the number of convolution kernels in the outermost layer of the generator.Experiments show that when the mask threshold is 0.45,the similarity of the reconstructed fingerprint is 86.32%,which is an increase of 12.97% compared with that before the improvement.The PSNR was 18.243,up 1.428.The SSIM is improved from 0.875 to 0.908,and the model improves the reconstruction effect of large incomplete fingerprint images to a certain extent.(2)Aiming at the problems of local fingerprint ridge blur,fracture and jaggedness in a few generated fingerprint images,this thesis improves a multi-scale Pix2 Pix network model to enhance fingerprint images.Through this model,the local fingerprint ridge lines are made smoother and clearer.The results show that when the mask threshold is 0.45,the similarity of fingerprint reconstruction is 89.64%,an increase of 3.32% compared with that before the enhancement.PSNR was 18.884,up 0.641,and SSIM was 0.913.The model further improves the quality and realism of the generated fingerprint images.(3)Aiming at the incomplete fingerprint image reconstruction model,this thesis develops a Java Web-based incomplete fingerprint image reconstruction system.The purpose is to simplify the user’s operation and provide a more friendly interface to make the user more aware of the process of reconstructing the incomplete fingerprint image.In summary,this thesis uses 76,000 fingerprint images for model training.Through the above two algorithm models,the reconstruction effect of incomplete fingerprint images is of higher quality and more authenticity than the images generated by traditional reconstruction methods.