To make the inpainted images better meet human eye visual requirement

this paper proposes a MRF image inpainting algorithm based on structure offset statistics and multi-direction features. On one hand

to better maintain structure coherence of the inpainted images

the degraded image is partitioned into structure and non-structure parts according to the edge features extracted by Curvelet transform

and the offsets between the similar patches are respectively counted. For each part a few dominant offsets are selected as the candidate labels according to their statistics. On the other hand

to better maintain neighborhood consistence between adjacent pixels

the energy function is constructed by incorporating multi-direction features. Experimental results show that the proposed approach outperforms several state-of-the-art methods.