Digital Image Inpainting Based On Partial Differential Equation

The coming of the digital age has promoted the rapid development of imageinpainting technology. The damaged areas of images are maximum inpainted by theknown information for constantly calculus and fill. This technique has very importantsignificance in cultural relics, film and television media, aviation, and so on. In recentyears, many outstanding scholars have proposed many algorithms, which promote thedevelopment of the inpainting technology in some extent. Using partial differentialequation model has attracted much attention because of the large inpainting advantageand good effects.Now three typical PDE-based inpainting models are being in-depth study,including high order partial differential equation BSCB, CDD model, and TV modelbased on variational functional, and then, simulate and verify the experiment resultrespectively for each algorithm. The thesis emphatically discusses the variational TVmodel. In view of disconnected vision defects and slow repair speed, the mainimprovement work is discussed by the following aspects:Firstly, the gradient of the traditional TV model usually artificially set a fixedvalue, without considering the influence of parameter selection for image inpaintingeffect and efficiency. According to the analysis, in early iterations, if the value isbigger which can more conducive to rapid spread of information; And at the end ofthe iteration, the numerical decrease is favorable for the formation of sharp boundaries,which can avoid the occurrence of false edges. At the same time, an integratedadaptive threshold method is put forward, which change the tradition algorithm thateach round of iteration use the same threshold. In this paper we give big parameters abig threshold, as the value of reducing, make a low stopping threshold. Thisimprovement while raise the effect of image inpainting, accelerate the rate of imageinpainting. Secondly, the traditional integral variational algorithm use the knowninformation of the restored images in diffusion step by step, but neglect the differentinfluence of different reference point. Therefore, this article introduces the correlationcoefficient, according to the distances among reference points and target points todetermine the correlation degree, then put relevancy coefficient into iterative formulafor testing. The experiments show that this improvement will make the edge of imagetransit more natural.Through the experiment result, above two improvements can be seen to get thebetter effect in image inpainting. At the end of the paper make summary and prospects,and show that the digital image research has very good prospect and value.