Research On Digital Image Inpainting Algorithms

Image inpainting means to use the known information in an image filling the unknown area according to some property rules when an image is damaged or some ojects in an image needs to be removed. After the inpainted process, the inpainted image is as close as possible the original image or no people knows this inpainted image is inpainted. Human vision is the main means to receive information and the image information is the most important visual information people received. In the process of image transmission it will inevitably be damaged, so image inpainting is becoming more and more important, especially in nowadays when the rapid development of the Internet, sharing, transmission, exchanging information have become very frequent. Firstly this paper introduces the digital image restoration research background and significance. Secondly we come to introduce the main ways of the digital image inpainting which contain partial differential equations, texture synthesis and sparse representation. Then we mainly focus on the image inpainting algorithm based on blocks of samples texture synthesis method and image inpainting based on sparse representation which is according to the image blocks.We find that the classical image inpainting based on block exemplar has the block effecting when we deeply study the block exemplar image inpainting. A new improved algorithm has been proposed based on the classical block exemplar inpainting framework. The new algorithm mainly changed to using a new way to calculating the weighted value of the boundary pixels. The image gray entropy entry is brought to the center pixels weight calculation. When calculating the weighted value of the center we use addition instead of multiplying to solve the problem that the degree of confidence quickly drops to zero with the repair process going on, which resulting in the wrong filling order and then resulting in the wrong inpainted image. The search area of the best matching block is reduced to a certain neighborhood to reduce computation time. Thinking of the non-local similar patches having the similar construct performance, combining with the gaussian mixture model, a new inpainting method based on non-local similarity blocks sparse representation is mentioned. This new method takes good use of the structural characteristics of these similar blocks.Image quality evaluation of the inpainted image is critical when comparing what kind of methods is better. This paper mainly from two aspects to evaluate the inpainted image:subjective and objective evaluation. Different experiments use different evaluation methods. Experimental results show that the improved algorithm is better with the block effect smaller and more in line with the aesthetic standards of human compared to the traditional digital image inpainting based on block exemplar. Similarly, based on non-local similarity blocks sparse representation combined with the gaussian mixture model image restoration method is much better than the restoration method based on image blocks sparse representation.