Research On Exemplar-based Color Image Inpainting Method In The Wavelet Domain

Digital image processing is a technology for denoising,enhancing,inpainting,and segmenting images through a computer,among which image inpainting technology occupies an important position and is a hot research field.Image inpainting can repair the damaged area according to the effective information in the damaged image,making the damaged image more coordinated on the whole.This thesis summarizes the principle of exemplar-based digital image inpainting and analyzes the characteristics of the existing image inpainting model and algorithm in the wavelet transform domain.On the basis of these,several improved methods of the exemplar-based inpainting model in the wavelet transform domain are proposed,which improve the image repair effect and repair efficiency.The main research work of this thesis are as follows:1.To overcome the drawback of the existing image inpainting algorithms for large-scale damage and strong structural texture,a novel exemplar-based color image inpainting algorithm is proposed,which uses the structural similarity and the characteristics of the wavelet transform domain.The proposed algorithm is a hierarchical classification method.At first,the damaged image is decomposed into low frequency sub-images and high frequency sub-images with different resolutions using the wavelet transformation.Then the high frequency sub-images which reflect the edge and texture information of the image,are repaired by an improved exemplar-based method,in which the best matching criterion is redefined by introducing the structural similarity(SSIM),and the priority is improved by using the wavelet transform coefficient.The low frequency sub-images which reflect the structural information of the image,are reconstructed by fast marching method(FMM)algorithm.Finally,wavelet reconstruction is used to complete the entire damaged image repair.Simulation results show that the image repaired by this algorithm has both a more natural visual effect and a higher peak signal-to-noise ratio(PSNR).Compared with related algorithms,PSNR is improved by 0.6-2.5d B.2.To further improve the ability of the image inpainting algorithm in the wavelet domain to repair the edge structure and detail texture,an algorithm based on information entropy in the wavelet domain is proposed,which can effectively reduce the occurrence of color incongruence and texture blur in the process of text removal and large-scale damaged edge repair.To repair the high frequency sub-images,the information entropy,which can describe the structure information of sample patch,is introduced into the calculation of the best-matched sample patch,improving the accuracy of matching results.Meanwhile,in order to improve the repair efficiency and engineering applicability,a method using the information entropy similarity and the structural similarity is proposed,which can calculate the size of sample patch adaptively.The simulation results indicate that the algorithm has a good repair effect on the edge structure and detail texture,and can maintain the continuity of the edge and natural visual effect,and improves the repair quality.Compared with related algorithms,the peak signal-to-noise ratio(PSNR)is improved by 0.4-3.1d B.Meanwhile,and the repair time is shortened.