Edge Preserving Convolution-Based Image Inpainting

Image inpainting is the process of filling in damaged or missed regions in an image using information from outside these regions. One of the best techniques for this goal is the convolution-based approach. The convolution-based algorithms are very quick, but they blur edges and structures, and they cannot reconstruct the edges. In this paper, a novel convolution-based algorithm is proposed whose goal is to overcome this drawback, and it could reconstruct edges properly. The proposed method uses a new 3×3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3 \times 3$$\end{document} convolving mask designed in such a way that all structures in the image can be preserved. The convolving mask is computed from the multiplication of four masks which two of them consider edges in horizontal and vertical directions and two others consider two oblique directions. Finally, the computed mask is convolved by the boundary of missed regions until the damaged regions are reconstructed. The proposed algorithm has adequate speed and outperforms some well-known algorithms in complex and unrepeated edges and smooth regions. The proposed algorithm is iterative and very simple to implement. Experimental results confirm the effectiveness of the proposed algorithm.