Study On The Digital Image Inpainting Technology

Digital image restoration is one of the research concerns in image processing, and that is alsothe research hot in computer vision and computer graphics. Image restoration attain restoration’sgoal is using algorithm to repair image which possesses scratch or damage and remove charactersor something unwanted and finishing the processing step for many high level applications.The existing digital image restoration models are separated into two types including thegeometric inpainting and the texture synthesis. This paper analysis and discuss the computingessences and the corresponding meaning of PDE based computing pattern witch used most indigital image geometric features restoration, based on the comparison of pixel-value-driving patternand the gradient-driving pattern, the reason why apparently the latter is superior to the geometricfeatures inpainting of large-scale regions is stated, then proposes that the information transmittingdirection always points the outside of inpainted regions is crucial to the inpainting result. Finally animproved method of typical gradient-driving inpainting algorithm is proposed.Then, based on the typical experiments of image geometric features inpainting using thegradient-driving model, the information transmitting direction is unstable and the inpainting resultsare well if the information transmitting direction always points to the outside of inpainted regions.Then we propose a method of gradient-driving image inpainting that can keep informationtransmitting direction pointing outside of inpainted regions. This method can stabilize theinformation propagating direction, so its inpainting performance is more robust.Finally, based on the analysis and typical experiments of image geometric features inpaintingusing the PDE based inpainting model mentioned above, we can find that the inpaintingperformance is well, but there will be vague effect and inefficient if many texture or mixture oftexture and geometric features are included in large-scale inpainted regions. Then we propose aninpainting algorithm based on steady field model. Not only the inpainting efficiency is stronger, butalso inpainting effect is maintained.