An Improved morphological component analysis algorithm for Tangka image inpainting

This paper proposed a new image inpainting method based on morphological component analysis that is capable of filling in holes in overlapping texture and cartoon layers. Firstly, due to rich content and complex color of Tangka image, the imposition of a total variation penalty by conventional model may not be accurate and easy to produce staircase. To improve the performance of sparse-representation-based image decomposition, in this paper the concept of non-local means which explicitly exploits self-similarities is introduced. In addition to, using fast algorithm reduce effectively the calculation of not related pixel weights within area, so the complexity of algorithm is reduced. The novel model preserve the fine structure, details and texture and eliminate staircase simultaneously, which make the subsequent iteration is more effective. Secondly, in order to improve the performance of sparse representation based image restoration, the concept of an example patches-aided dictionary learning algorithm named KSVD algorithm is adopted. Experimental results for thangka image which contains scratch and block loss show that the proposed method achieves better inpainting effect.