Color-direction patch-sparsity-based image inpainting using multidirection features

This paper proposes a color-direction patch sparsity-based image in painting method to better maintain structure coherence, texture clarity, and neighborhood consistence of the in painted region of an image. The method uses super-wavelet transform to estimate the multi-direction features of a degraded image, and combines with color information to construct the weighted color-direction distance (WCDD) to measure the difference between two patches. Based on the WCDD, the color-direction structure sparsity is defined to obtain a more robust filling order and more suitable multiple candidate patches are searched. Then, the target patches are sparsely represented by the multiple candidate patches under neighborhood consistency constraints in both the color and the multi-direction spaces. Experimental results are presented to demonstrate the effectiveness of the proposed approach on tasks such as scratch removal, text removal, block removal, and object removal. The effects of super-wavelet transforms and direction features are also investigated.