an image inpainting method

An image inpainting method BianRu Li, Yue Qi, XuKun Shen School of Computer Science & Engineering, BeiHang University {libr, qy, xkshen}@vrlab.buaa.edu.cn

Abstract Image inpainting technique has been widely used for reconstructing damaged old photographs and removing unwanted objects from images. In this paper, we present an image inpainting method based on the existing exemplar-based image inpainting idea. Our method improves the robustness and effectiveness by rational confidence computing method, matching strategy and illing scheme. Thereore, our method eectively prevents “growing garbage”, which is a common problem in other methods. With our method, we can obtain preferable results to those obtained by other similar methods.

1. Introduction ĀInpainting” is an art world’s term borrowed from museum restoration artists. This activity consists of filling in the missing areas or modifying the damaged ones in a non-detectable way by an observer not familiar with the original images [1]. Applications of image inpainting range from restoration of photographs, films and paintings, to removal of occlusions, such as large unwanted regions, superimposed text, subtitles, stamps and publicity, from images. In addition, it is more significative to restore the precious calligraphies and paintings in the digital museum with image inpainting technique. Previously, the two typical image inpainting methods are texture synthesis [2], [3] and mathematically computational image inpainting [1], [4], [5]. In the former, as shown in figure 1(a), constrained texture synthesis has been used for filling in the missing information (in a hole) from the surrounding region. However, it is only applicable to the pure texture images and has difficulty in filling holes in images which consist of linear structures. In the latter, shown in figure 1(b), Bertalmio [1] pioneered a mathematically computational digital image inpainting

algorithm based on partial differential equations. It fills in the areas to be inpainted by propagating information from the outside of the masked region along isophote which is perpendicular to the gradient vector of the restored region contour. This intends to propagate information while preserving edges. The disadvantage of the method is that the diffusion process leads to some blur, which becomes noticeable when filling larger regions. Therefore, their algorithm best performs for pure structure images [6] and those with thin cracks and text.

(a)(b) Figure 1. Previous methods. (a) The result of texture synthesis [3]. (b) The result of mathematically computational image inpainting [1].Since most images are not composed of pure texture or pure structure, methods [6], [7], [8], [9] combine the advantages of both texture synthesis and mathematically computational image inpainting. Approch [6] decomposes the original image into structure and texture regions, and they are processed by mathematically computational inpainting and texture synthesis separately. The output image is the sum of the two processed components. This approach still remains limited to the removal of small image gaps, and the diffusion process continues to blur the filled region. Our method is based on the exemplar-based image inpainting algorithm [7]. It has the efficiency and qualitative performance of exemplar-based texture synthesis, but also respects the image constraints imposed by surrounding linear structures. Therefore, it

Ninth International Conference on Computer Aided Design and ComputerGraphics (CAD/CG 2005)0-7695-2473-7/05 $20.00 © 2005IEEE

Authorized licensed use limited to: Soo Chow University. Downloaded on April 21, 2009 at 04:34 from IEEE Xplore. Restrictions apply.is a simple region filling algorithm which does not suffer from blur artifacts. The results are impressive and compare favorably with those obtained by similar techniques [6], [8], [9]. However, this algorithm has some problems: firstly, it merely adopts a simple priority computing strategy without considering the accumulative matching errors; secondly, the matching algorithm for texture synthesis only uses the color information; thirdly, the filling scheme just depends on the priority disregarding the similarity. As a result of lacking robustness, their algorithm sometimes runs into difficulties and “grows garbage”. To solve these problems, we propose an image inpainting method. The structure of this paper is as follows. In Section 2, we describe our inapainting method in details. We show the practicability and effectiveness of our approach using a set of real and synthetic images in Section 3. We conclude in Section 4.