Research On Image Inpainting Based On Sample

Image inpaint ing is an important research content in t he co mputer graphics,it s goal is to repair the specified area,and has the requirement o f natural transit io n o n final image,which makes the observers not perceive the art ificial trace as much as possible.Image inpaint ing is widely used in do mains such as restorat io n o f video and damaged images,remo val o f barriers,inpaint ing o f ancient cult ural relic,the set of special effects,image zoo ming and video co mmunicat io n error concealment and so on.The recent ly research o f digital image inpaint ing techno log y mainly focus o n two aspects: image inpaint ing based o n part ial different ial and var iat io n equat io ns,and image inpaint ing based on texture synthesis.The former is based on PDE inpaint ing model,fills t he areas to be inpainted by smoothly propagat ing infor mat io n fro m t he known surrounding areas alo ng the normal o f direct io n iso phote,and it is applicable to sma ll area images.The latter is based on t he consist ency,co herence,repeatabilit y o f texture and makes use o f the effect ive synt hesis o f texture to co mp lete the inpaint ing,it is mainly suit able for the images wit h r ich structure texture and large scale damaged area.In t his paper,we studied several representat ive image inpaint ing mo dels,wit h the t horough analysis o n t he algorit hm o f Criminisi,we make a co mpariso n and analysis on visio n qualit y,inpaint ing efficiency,peak signal to no ise rat io and t hen make two impro vement s for it s existed insufficiencies.Consider ing t hat the tradt io nal repairing algorithm based o n sample can eas ily causes the disco nnect io ng and blurring during reparing t he strong structure images.We introduce a part ial different ial equat io n smoothness constraint term and using new matching crit erio n to optimize t he order o f inpaint ing,which realizes t he inpaint ing o f large area images w it h t he help of PDE.Consider ing t he pro blem t hat the unreaso nable design o f priorit y equat io n o n Criminisi,and it may generate so me redundant pixels in the process o f image inpaint ing,which will lead to error accumu latio n,and the size o f matching block is subjected to the co mp lexit y extent o f texture and so on,an impro ved exemplar-based met hod for image inpaint ing is proposed.We mainly make so me impro vements as fo llo ws: taking into account the curvature term and average gradient term,which have an important impact on t he select io n o f final samp le blo ck,we redefine the data term;Considering the dynamic changes o f curvature distance in the process o f inpaint ing,which will avo id error accumulat ion problem,we redefine the confidence term;Changing the order o f inpaint ing when dealing wit h so me images wit hout strong edges,which will get better result s.A large nu mber o f experiment al resu lt s show that the algorit hm we proposed can get more sat isfying effect in the aspects o f visio n qualit y and inpaint ing efficiency.