Research And Implement On The Improved Algorithms Of Image Inpainting

Image inpainting aims to restore the missing part of an image through the information of the source region, make the image more naturally with out repair evidence after restoration. Originally, this technology use manual fill to restoring the cultural relics of the old days. Nowdays, digital image inpainting following the manual fill ideas, and start to fill the missing part of an image by the computer and other auxiliary tools, it overcomes the risk of manual restoration. This technology has widely used in many fields, like industry, communication, medicine, film, television and so on.There have two categories in the digital image inpainting methods: one is based on partial differential equations which mainly for small missing part restoration, and the other is based on texture synthesis that suits for filling large missing areas. In this paper, we will introduce the two inpainting methods’algorithm models and the inpainting results for details. In order to improve the repair quality image, we make some improvements at the Criminisi algorithm and take many experiments to prove their feasibility.1) Find out a fixed proportion as the weight ration in priority function, here the function is the addition of the data term and the confidence. We compared two sets of images with different weight ratio, finally using the golden section as the weight ratio, and experiments shows that our improved algorithm 1 avoids the weight ratio’s select when restoring different images, and make the confidence drop to zero more slower.2) The confidence drop to zero rapidly in the inpainting process, even our improved algorithm 1 make its drop speed slower, but with the increase of iterations, the confidence still tends to zero. We introduce the logistic equation in the confidence updating function to overcome the limitation of our improved algorithm 1. Our improved algorithm 2 was proposed after the selection of the parameter, here the confidence can automatically adjust in the inpainting process, and also makes a better inpainting results.We combine these two improved algorithm together at our experiment, and let the system automatic select the corresponding algorithm after compute the missing area. Experiments show that our improved algorithms can reduce the error propagation and obtain a better restore effect than that of the Criminisi algorithm. However, the restore time costs is a little more than that of Criminisi algorithm, but our improved algorithms can increase the visual effect of the inpainting image and have a high quality inpainting image results.