TV Model Image Restoration Technology Research And Evolution Algorithm

With the development and improvement of electronic products and communicationtechnologies, more and more digital images appear in people’s daily life. While in reality, manyman-made or objective factors will cause images damaged. For a simple and effective way torecover the damaged images that have a certain value and significance, more and moreresearchers constantly study in digital image completion techniques. In recent years, digital imagecompletion has become a popular research topic in the fields of computer graphics and computervision, and has a very widely range of applications.At present, digital image completion techniques are mainly divided into two categories: Oneis image inpainting technology based on variation PDE; the other is image completion techniquebased on texture synthesis. This paper mainly discuss the classic algorithms of the two categoriestechnology above-mentioned: the image inpainting algorithm based on TV model and texturesynthesis image completion based on sample. On the basis of analyze and study the tworepresentative algorithms, two new algorithms are proposed in this paper. The main contents andinnovations are as follows:(1) An improved image inpainting algorithm based on TV model is proposed. In order to usemore information of known pixels closely related to the damaged pixel, this proposed algorithmuses distance of two aspects to confirm diffusion factor of neighboring pixels. One is the pixelnumber between the damaged pixel and its neighborhood pixels; the other is the distance ofdamaged pixel to the isophote of its neighboring pixels. With little pixel gap and short isophotedistance, the neighboring pixels own big diffusion factor. This algorithm can effectively minimizethe error accumulation caused due to using the known information with less relevance.(2) A new priority function is employed to calculate the priority of damaged patches.According to some deficiency in calculating the patch priority in texture synthesis imagecompletion based on sample, this paper redefines the priority function. In the new function, thestructure information of damaged patches occupies greater proportion. The data item has beenimproved, so that image completion considers the information both on isophote and gradientdirection at the same time.(3) A search strategy based on evolutionary algorithm is applied to search the optimalmatched patch. In texture synthesis image completion based on sample, the efficiency ofsearching matched patches is low and it’s easy to bring mismatch. So this paper use evolutionaryalgorithm to search the most similar matched patch in the ring area around the damaged region.And in the whole process of searching, when the gradient of the center pixel of damaged patch islarge, the size of template window is small, vice versa.