Research On Large Area Image Inpainting And Special Effects Generating

Digital image inpainting mainly aims at filling in missing pixels in an unknown region of an image in a visually plausible way. And it is also one of the important research fields in computer graphics, computer vision and image processing. Currently, digital inpainting techniques have found broad applications in image processing, vision analysis, digital restoration of ancient paintings for conservation purposes, text removal and objects removal in images for special effects, restoration of old photographs or films with scratches or missing patches, disocclusion in computer vision, errors conceal in videos, and so on. The primary methods of image inpainting include"Texture Synthesis" algorithms and the algorithms based on Partial Differential Equations (PDEs). This thesis fulfilled the image inpainting algorithm based on radial basis functions. On the basis of examplar-based inpainting, this thesis carries out some improvements to achieve the large area filling. In the end, we established a special effects producing system by using its relative applications.In the concept of algorithm, we found that the exemplar-based inpainting method searching the source patch globally easily lead to some matching errors, and such errors would accumulate and propagate into other areas quickly, even making the results completely unacceptable. In this paper we use local part searching, priority improving, and Graph Cut method which are coming from the original algorithm to solve the problems that would occur when we are dealing with large area's filling.(1) One of the innovations of this paper is the local partition searching strategy, we made a new local searching strategy, which can identify the search area according to the property of the known pixels that are belonged to the currently filling area, so as to reduce the false matching which are brought by the global searching effectively.(2) In order to guarantee that the structure component of the image can be filled first and resolve the problem of expanding from the high texture area to the low texture area, we discuss and infer the formula of the priority. The improvement of the formula of the priority is for a better guidance to the new searching strategy.(3) Copying pixels directly will cause a visual inconsistency between the original part and the filling part. Using Graph Cut, we can calculate the most optimized sewing ling, which will make the transition of the fixing blocks looks more natural on visual.Experimental results show that this improved method can efficiently handle different cases, especially the large regions in complex background.In the concept of system implementation, we apply image cutting method during object extraction process, then generate a new background by using large area image inpainting method to achieve the goal of recognizing and filling an area of an image automatically.On the basis of previous work in our lab, we present a digital image special effect production system which is designed for ordinary users. This system contains the functions of object extraction, image fixing and image synthesis etc. Experiment shows the system's usability and efficiency, and it can give us satisfactory results.