An Improved Criminisi Algorithm For Image Inpainting

As the saying goes: one figure is worth more than thousand words.It means that the images are small and simple,but form and shape of image varies.A small image offen carrys a large amount of information,and the words are too raw and abstract to understand,and in recent years,along with the development of network technology and the wide application of digital multimedia technology,the digital images can easily be found anywhere in our daily life.But in the process of digital image storage and transmission,some of local information of image may be damaged,and in some other areas,such as protection of cultural relics,conceal errors,removing obstacles,special effects,etc,are also need to repair the image.So the importance of the study of image restoration algorithms are obvious.The goal of digital image inpainting is to fill the region where information is missing or replace with different image making the image look natural,smooth and no obvious trace of manual processed.The digital image inpainting technology is mainly divided into two branches.One is based on partial differential equations(PDE),which fills the areas to be inpainted by smoothly propagating information from the known surrounding areas along the normal of direction isophote based on the principle of diffusion in pixels,thus complete the whole inpainting.for narrow and non-texture regions the method has a very good repair effect.Another is a method of exemplar-based image inpainting this method is based on pixel block.This method replaces the target block with sample which is similar to target block in intact area with a certain filling order and then completing the image inpainting.This method is suitable for damaged images rich in texture.This paper further analyzes several typical digital image restoration model,and simulates the Criminisi algorithms which are mproved by relevant researchers in recent years.We found that it have many shortages in vision quality and inpainting efficiency.So it is improved in this paper from the following three aspects.In the image inpainting process,the repairing order and the matching accuracy of the Criminisi algorithm is not reasonable enough between target block and sample block.Aimed at these shortcomings,an improved algorithm based on the Criminisi algorithm is proposed in this paper.The Local Entropy of image is added to to measure the size of the known information in target block when calculating theconfidence term which make the priority computation more reasonable.On the matching criterion we adopt a mehtod of the coarse-grained and fine-grained combination.First introduced matrix similarity criterion of coarse-grained block matching,shorten matching time,and then use the fine-grained pixel matching,improve the matching accuracy.