The Techiques Of Digital Image Inpainting

Image inpainting is an image processing, which fills in damaged portions of a given image. The technique of image inpainting is used widely in many fields such as repairing of damaged medical image, ancient artifact and removing of scratches and spots of dust on in film etc.This dissertation is mainly focused on the methods of image inpainting. At first, we introduce the background, the development conditions of the image inpainting and its search conditions, then summarize the basic knowledge of the technique, classification of image inpainting and several typical image inpainting models. At last, we introduce two methods, which include a novel image inpainting algorithm based on neighbour pixels and an algorithm of completion for strong orientational texture images.Currently, the main methods for image inpainting include "texture synthesis" algorithms and the algorithms based on partial differential equations (PDEs), which are either difficult to understand, or complicated to implement. In this dissertation, by defining the priority of inpainted points, we present an image inpainting algorithm based on neighbour pixels of the points. Our algorithm avoids both the expensive computations of texture matching and the time-consuming solutions of complicated PDEs, and therefore is easy to understand and implement.Natural photographs and strong orientational texture images sometimes may contain stains or undesired objects covering significant portions of the images, where the removal of a foreground object creates holes in the images. We present an inpainting algorithm for strong orientational texture/color images existing in our environment to fill in the holes. Becauce of the strong orientation, the algorithm synthesizes the missing parts by image patches selected from orientationally local areas. To search along the orientation of texture /color distribution or the constrained orientation, we select a source block BTN near a missing block BT firstly, and find the most similar block BSN to BTN, then update BT by the block BS near BSN. To reduce blockiness, the boundary between blocks is further computed as a, minimum cost path through the error surface at the overlap before updating BT. A few examples for a variety of texture images are given to demonstrate the effectiveness of our method.