###
DOI:
Journal of Software:2006.17(zk):138-147

基于大位移视点图像的单帧图像修复
刘春晓,潘梁,郭延文,王进,陈为,彭群生
(浙江大学 CAD&CG国家重点实验室,浙江 杭州 310027;浙江大学 理学院 数学系,浙江 杭州 310027;浙江大学 CAD&CG国家重点实验室,浙江 杭州 310027;南京大学 计算机软件新技术国家重点实验室,江苏 南京 210093)
Image Inpainting Based on Large Displacement View Images
LIU Chun-Xiao,PAN Liang,GUO Yan-Wen,WANG Jin,CHEN Wei,PENG Qun-Sheng
()
Abstract
Chart / table
Reference
Similar Articles
Article :Browse 2486   Download 2535
Received:March 15, 2006    Revised:September 11, 2006
> 中文摘要: 提出一种基于大位移视点图像的单帧图像修复算法,利用大位移视点图像中的可见信息修补目标图像中的被遮挡或信息丢失区域.算法的关键在于如何转化大位移视点图像的可见信息为可用信息,以及如何利用得到的可用信息来有效地修补目标图像.在交互指定待修复的目标区域后,算法首先将所有图像分割为不同的平面场景区域,并基于图像匹配将大位移视点图像中的平面场景区域变换到当前视点.因此,其中的可见信息就可被直接使用.进而通过定义合适的修复和融合优先级函数,提出基于纹理合成和图像融合的图像修复算法,利用获得的可用信息来修补目标区域.修复区域和目标图像之间的鬼影现象使用Poisson图像融合算法来消除,以达到无缝的修复结果.实验结果表明,该算法能够修复较大的丢失信息区域中的结构和纹理信息,具有一定的实用价值.
Abstract:This paper presents an algorithm for image inpainting based on the large displacement view images, it fills in the occluded or damaged regions of the target image using the visible information from other large displacement view images. The key issues are, first, how to convert the visible information in the large displacement view images into useful information; second, how to make use of the useful information to repair the target image effectively. After specifying the target region with interaction, this paper tackles the first issue by dividing all the images into different planar scene regions, and then transforming all the regions in the large displacement view images into the current view by image matching. As a result, the visible information in them can be directly used. For the second issue, through the properly defined repairing and fusion priority functions, a new image inpainting algorithm is developed based on texture synthesis and image fusion to repair the target region with the acquired usable information. In addition, the ghost effect between the repaired region and the target image is eliminated by Poisson image fusion to generate the seamless result. Experimental results show that this algorithm can recover the structural and texture information in the large missing region, and holds its practical value.
文章编号:     中图分类号:    文献标志码:
基金项目:Supported by the National Natural Science Foundation of China under Grant No.60403038 (国家自然科学基金); the National Grand Fundamental Research 973 Program of China under Grant No.2002CB312101 (国家重点基础研究发展规划(973)) Supported by the National Natural Science Foundation of China under Grant No.60403038 (国家自然科学基金); the National Grand Fundamental Research 973 Program of China under Grant No.2002CB312101 (国家重点基础研究发展规划(973))
Foundation items:
Reference text:

刘春晓,潘 梁,郭延文,王 进,陈 为,彭群生.基于大位移视点图像的单帧图像修复.软件学报,2006,17(zk):138-147

LIU Chun-Xiao,PAN Liang,GUO Yan-Wen,WANG Jin,CHEN Wei,PENG Qun-Sheng.Image Inpainting Based on Large Displacement View Images.Journal of Software,2006,17(zk):138-147