Research on Mural Inpainting Method based on MCA Image Decomposition

Morphological component analysis (MCA) is a signal analysis method based on sparse model, its core idea is to represent different components of the signal based on morphological difference of the signal's components. It can separate overlapping texture and cartoon image layers by use two adapted dictionaries. MCA performs good in image inpainting, especially for the image scratch repairing, small area filling, and remove small object. In this paper, we proposed a color image inpainting algorithm based on MCA, and applied the proposed algorithm to repair the murals digital image in the Shibaoshan grotto of Jianchuan, Yunnan province. The central idea of the algorithm is converting the color image into the Lab color space, and inpainting the texture and piecewise smooth (cartoon) parts respectively. Meanwhile, our method increases the TV penalty term in the sparse representation of the cartoon parts to reduce the effects of the noise. Finally, the method combines three channels to realize the color image restoration. Experimential results show that the method has good performence on scratches and cracks in the inpainting of digital images of mural paintings.