Image Inpainting, (generally known as image completion), is the technique to fill holes in an image. The challenge of image inpainting is how to fill the hole in a visually plausible way. The success of structure propagation is highly dependent on the order in which the filling proceeds. Hence, the authors have proposed a computationally efficient algorithm by a patch-based sampling process. Although the algorithm propagates the structure well and produces some amazing results, it still has difficulties when inpainting images where complex salient structures exist in the missing regions. Therefore, the user is allowed to manually specify the important missing structure information by extending a few curves or line segments from the known to the unknown regions. The curves or line segments would be treated as the constraints, and then the structure propagation is formulated as a global optimization problem.

In this project, we implemented two inpainting algorithms. The first one is exemplar-based inpainting, and the second one is an extension to the first, which allows user to manually specify the structure information of the region to be inpainted. The algorithms are rather robust in the sense that the parameters, such as weighting factors for the data term in structure propagation and patch size, are insensitive. We do not have to spend much effort on tuning parameters but still can get good results. Generally speaking, current inpainting algorithms work well when region to be inpainted has complex texture, such as grass, cloud, and waves, or when object to be inpainted is natural. However, they may fail when we try to inpaint artificial things, like human bodies and faces. Also, these inpainting algorithms have their limitations. The most common problem is that when there is no suitable patch in the same image, the algorithms may not be able to produce promising result. One possible solution is to develop an algorithm that can automatically retrieve suitable sample images from the Internet, and complete inpainting with such samples.

Main focus – Exemplar based, Image Inpainting, Structure Propagation, Onion Peeling Method