Large Damaged Area Image Inpainting Algorithm Based On Broken Structure Line Matching Model

In this paper, the main research is large damaged region image inpainting algorithm. To solve these problems, such as mismatching connecting or unsmooth connecting when inpainting large damaged region with complicated structure information, according to the idea from coarse to fine, the broken structure line is connected at first, and then inpaint the sub-regions which are divided into different part by the connected structure lines, an image inpainting algorithm based on matching model for broken structure line is proposed in this paper. The main work is as follows:1) Firstly, several potential factors which impacts the calculation of the matching degree between the broken structure lines are analyzed deeply and researched the relationship between them, different weights are assigned into different factors based on their significance. On that basis, a matching model for broken structure lines is constructed to get matching pairs among these broken structure lines. Secondly, according the matching pairs of the broken structure lines, it can be connect with the curve fitting. Meanwhile, speculate the possible connection, and using the different inpainting method for the different connection, and the inpainting result of structure lines are nice in aspect of visual connectivity. Finally, we using the block matching algorithms inpaint the area which is divided into different sub-region by the connected structure lines, namely, and in order to get better repair effect, limited the search area according to the location of the pixel which need to inpaint when searching for the best matched block.2) In order to make full use of the structure information that has been inpainted, in the third section; we used the efficient belief propagation via priority scheduling and dynamic pruning algorithm which proposed by Komodakis to inpaint the damaged sub-region. An adaptive priority BP algorithm is proposed to solve the problem that set the candidate label parameter is manually in the original p-BP algorithm. By calculating the constraints of the overall matching degree of the label set and known information, which can adaptive definite threshold of the candidate label, and then adaptively determine the priority of the node and the threshold of the label selection.Experimental result shows that the proposed algorithm is better than others in objective evaluation indicators about the PSNR and SSIM, it is able to obtain the desirable result, and is also easy to understand and verification.