A Simplified Image Inpainting Method Based On Surface Shape Approximation

Digital image inpainting is a technique for repairing a partially damaged or lost image in an undetectable way. It fills the damaged part of an image using information in the undamaged neighborhood according to certain rules so that the repaired image looks a natural one. Image inpainting technique has been widely used in various fields, including heritage restoration, crime scene reconstruction, medical image repairing, film scratched and stains removal, and so on.From the mathematical view, inpainting is clearly an ill-posed problem due to lack of sufficient information to guarantee the existence and uniqueness of a solution. For this reason, researchers have proposed various approaches to solve the problem with assumptions from visual psychology and perception as constraints.In this thesis the background, present status and existing problems of the image inpainting research are first addressed, followed by an introduction to knowledge of visual psychology and a summary of the basic principles in this field. Then two classical structure-based inpainting algorithms are discussed and implemented. They are 1) Bertalmio's method of propagation along the direction of level line (BSCB) and Chan's algorithm base on total variation (TV) model. Then we proposes a simplified inpainting method based on surface shape approximation, which takes damaged image as a damaged surface, and turns the image inpainting to the surface approximation to shapes. The experiment results show that this method well works in the case of the damaged part is long and narrow, or is surrounded by slow varying neighborhood. Finally, a summary and the prospects of future research are given.