Research Of Wavelet Image Inpainting Method In Pixel Domain

Image inpainting has been one of the important contents of digital imageprocessing. It aims for restoring image to satisfy human vision request through fillingthose losing parts of the damaged images using residual information.Wavelet transform which is characterized by analyzing signal in time-frequencyfield has been a new tool to analyze signals and images. During image processing usingwavelet, for some reasons some of wavelet coefficients are lost which results indamaged images in pixel domain after reconstruction. Wavelet image inpaintingproblem was firstly introduced by Chan et al. and later the method based on non-localtotal variation was proposed by Zhou and Chan. However, compared with imageinpainting in pixel domain, wavelet image inpainting is more complex and developsmore slowly.The thesis briefly reviewed the development of digital image inpainting, andintroduced wavelet analysis in details, then reviewed wavelet image inpainting. On thebasis of the research on the traditional methods of wavelet image inpainting, the ideathat wavelet image inpainting solutions are not constrained by wavelet domain can bein pixel domain.Based on such an idea ROF model, non-local means, dictionary learning based ona beta process were introduced, some methods in pixel domain to solve wavelet imageinpainting were proved to be effective by image inpainting simulating experiments. Theresult showed that connectivity principle was well realized during using our newproposed method to solve wavelet image inpainting in pixel domain and the PSNR ofdamaged images was greatly improved with good vision effect.