基于纹理规则性分析图像填补方法.docVIP

基于纹理规则性分析的图像填补方法* 郝川艳1+ 陈昕2+ 吴恩华3 1（福建师范大学协和学院 信息技术系，福建福州 350108） 2（福建师范大学协和学院 信息技术系，福建福州 350108） 3（澳门大学科技学院，计算机与信息科学系，澳门 999078） 摘 要：本文提出了一种利用纹理规则性分析和纹理合成技术进行高效图像填充的新方法。该方法可以自动地抽取图像前景区域并且合理的复原被遮挡区域的原始内容。对于前景抽取问题，一般的做法为了得到准确的结果往往需要引入交互操作。为了实现自动化的抽取，本文着重考虑前景为近似规则纹理（NRT）的场景区域。近似规则纹理具有强烈的信号规则性，利用对这种规则性的分析可以有效而准确的分割前景部分。对于背景恢复问题，我们基于纹理合成技术提出了一个新的背景填补算法，并且通过引入合成优先度概念将图像修补算法结合进来。因此该方法具有两种方法共同的优点。 关键词：图像填补; 纹理规则性; 纹理合成; 近似规则纹理; 图像绘制; 前景分割; 填充 An Efficient Image Editing Method by Texture Regularity and Texture Synthesis Hao Chuanyan1+, Chen Xin2+, Wu Enhua3 1(Department of Information Technology, Concord University College, FuJian Normal University, FuZhou, 350108, China) 2(Department of Information Technology, Concord University College, FuJian Normal University, FuZhou, 350108, China) 3(Department of Computer and Information Science, Faculty of Science and Technology, Macau University, Macau, 999078, China) + Corresponding author: HaoChuanyan Phn: +86 Fax: +86-**-****-****, E-mail: waterwu_ zz@ , http://**** + Corresponding author: ChenXin Phn: +86 Fax: +86-**-****-****, E-mail: c_risine@ tom. com, http://**** Abstract: This paper presents a new method for image completion based on texture regularity analysis. It extracts foreground information within image, fully and automatically for certain scenes, and fills the overlapping regions in a visually plausible way. Compared with previous works that usually select foreground objects interactively for a robust extraction, our work focuses on a special kind of foreground content called near-regular textures(NRT). Near-regular texture has strong regularity for various features that can be used to ease the automatically foreground segmentation and to help improving precision. In filling phase, we propose a combined framework by taking advantages both of “texture synthesis” and “inpainting”. This framework reconstructs the background by introducing priority into exemplar-based texture synthesis.