Research On The Techniques Of Digital Image Inpainting

In this thesis firstly we describe the background, theories and some typicalinpainting algorithms of digital image inpainting.Secondly we study the TV model in detail. However this method istime-consuming and has bad restoring effect caused by the inpainting order of lineby line, and the adoption of much information of damaged area in the process. Wegive a fast algorithm based on the TV model which takes the outside-in inpaintingorder and uses the information of the undamaged or restored area for imageinpainting. In addition, the method carries out only one time to finish the inpaintingwhile the TV model needs numerous times of iteration to do that. The results ofexperiments show that the algorithm we give in this thesis can reach much betterrestoring effect with much less computation time compared with the TV model.Lastly, since the TV model does not meet the connectivity principle, weanalyze the reasons, and give an improved fast algorithm. This method takes theoutside-in inpainting order and completely uses the information of the undamagedor restored area for image inpainting, and uses different methods to repair the edgepoints and non-edge point of the image. The results of experiments show that thisalgorithm meet the connectivity principle well, and can reach much better restoringeffect compared with the TV model and the fast TV model.