Research Of Image Inpainting Algorithm Based On Isophotes Computation

The traditional inpainting methods for small scale damaged images based on variation and PDE(Partial Differential Equations) is effective but time-consuming. Recently, With the development of algorithms based on isophotes computation in analysis of local characristics, the efficiency of inpainting is increased to some extent. But there are still some shortcomings. In order to inpaint the images with complex texture, the images can be layered to structure and texture by means of decomposition models, so as to combine the isophotes computation for structure inpainting and other algorithms for texture inpainting to get better visual effects.Firstly, the research’s current situation and significance of digital image inpainting technology are summarized in this paper, and the theory, principle and characteristic are explained in mathematical terms and visual psychological terms.Secondly, FMM(Fast Matching Method) based on isophotes computation is studied. And four rectangular inpainting templates for the damaged images caused by packet loss in network transmission is given in analysis of the FMM’s realized characteristics. In the meantime a certain inpainting effect for images with small rectangular damaged areas is proved by simulation experiments.Thirdly, CIIC(Convolution Inpainting based on Isophote Computation) is studied here, which combines the isophotes estimated algorithm and the convolution inpainting method. To express the multidirectional images better and select the convolution smooth templates correctly, an isophotes estimation using modified soble is proposed, and an allowed tolerance of the isophotes’direction is given. The CIIC using modified sobel inpaints the small scale damaged images better and faster, along with experiments to prove it works.Lastly, VO decomposition model is chosen by comparing the performances of DCT and VO. On the basis of layered images using VO, the isophotes computation for structure inpainting and DCT for texture inpainting is used in combination for images with wider and more complex areas lost. Through simulation experiments, the feasibility is proved and better results are obtained than CIIC.