Large Scale Image Inpainting Based On Exemplar And Structure Information

Image inpainting mainly aims at filling in missing pixels in an unknown region of animage in a visually plausible way. It is a hot spot in the research of computer graphics,computer vision and image processing. Image inpainting has been wildly applied in an-cient paintings for conservation purposes, old pictures'repair, text and objects removal inimages, errors conceal in videos, digital zooming etc.Currently, the image inpainting technology can be roughly divided into two cat-egories, PDE-based inpainting algorithm and texture-based inpainting algorithm. Theformer algorithm can achieve a good result when the missing region with small scale.However, the result is poor when the image has much texture information and missingregions with large scale. The latter algorithm can achieve a good result with the case oflarge missing regions. The result is poor when the image has too much structure informa-tion. Since images have different texture and structure information, this paper introducesan algorithm of large scale image inpainting based on exemplar and structure informa-tion. If images have rich texture information and less structure information, an improvedalgorithm of image inpainting based on exemplar is provided. If images are rich in bothtexture and structure information, an algorithm which is priority to repair structural infor-mation is provided. Images often contain straight line structure and curve structure, butthe classic algorithm has a bad result when inpainting this kind of images. This paper in-troduces an image inpainting algorithm based on straight line structure information and animage inpainting algorithm based on curve structure information. The image inpaintingalgorithm based on straight line structure information firstly utilizes Hough transform todetect structure features, then inpaints the straight line part, and finally uses the improvedalgorithm of image inpainting based on exemplar. The image inpainting algorithm basedon curve structure information firstly uses the way of image segmentation to find theobviously curve feature, and utilizes the least square method to fit a new curve. Theninpaints the curve part, and finally uses the improved algorithm of image inpainting basedon exemplar.Experimental results illustrate that the proposed algorithm of large scale image in-painting based on exemplar and structure information has a satisfactory result, and the algorithm is simple and easy to implement.