Comparative analysis of exemplar based image inpainting techniques

摘要

Inpainting has been an attractive and useful art in this fast growing era of digital image processing. Its applications in numerous fields such as restoration, object removal, superimposed text removal, red-eye removal etc. has made an impression on researchers to evolve generation of inpainting methods. After introduction of exemplar-based algorithm, several variations have been made in order to improve quality of inpainting region containing structure and texture information. Since inpainting starts from the border of the structure and propagates inward, selection of appropriate points on the structure becomes a very important criterion. Data term is responsible for extracting the structures in an image. With overview of exemplar algorithm, few variations in gradient based data terms are presented. Moreover, a data term based on sparsity of the image structure can also be an area of research. Evaluation of all these techniques is done using common analytical measures such as Peak Signal to Noise Ratio (PSNR), Similarity Structure Index Measure (SSIM) and Feature Similarity Index Measure (FSIM) in this paper. Besides qualitative comparison, subjective evaluation using Mean Opinion Score (MOS) is also presented.