Enhanced Algorithm for Exemplar-Based Image Inpainting

Image in painting is a method to fill in missing region using the information from the remaining area of the image in a visually plausible way. Exemplar-based in painting methods iteratively search the source region and fill the missing region with the best matched patch obtained by SSD criteria. But sometimes matching error would be easily generated because the SSD criteria do not consider the relationship of candidate patch and current patch in damage region. In this paper, we propose an improved criteria to select the best fit candidate using variance. The proposed method compares the variance between the current patch and the candidate patches to select the best matched patch whose variance is smallest on the base of SSD criteria for preventing error accumulation and propagation. Experimental results show a significant improvement on SSD criteria in both visual and mathematical aspects.