基于曲率和中值滤波的快速推移图像修复方法  

Fast marching image inpainting method based on curvature-driven and median filter

作　　者：张平[1] 檀结庆[1] 何蕾[1] 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009

出　　处：《计算机应用研究》2007年第8期186-188,192,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(10171026;60473114);安徽省自然科学基金资助项目(03046102);安徽省教育厅科技创新团队基金资助项目(2005TD03)

摘　　要：提出了一种结合中值滤波和基于曲率扩散的图像修复方法。中值滤波修复方法对于灰度值变化比较平缓的图像,修复效果从时间性能和视觉效果上均比较好,但对于灰度值变化比较剧烈的图像,效果不能令人满意。在结合了基于曲率扩散的修复方法之后,中值滤波的不足之处得到了改善。基于曲率扩散的修复方法考虑了图像本身的几何信息,在修复图像的几何特征上有较高的准确度。图像修复顺序由已知区域向未知区域扩散时经过各破损点的时间决定,用快速推移方法计算该时间。实验表明,该算法能高效、稳定地处理破损区域的图像信息。This paper introduced an algorithm that was the combination of median filter algorithm and the curvature-driven diffusion algorithm. The inpainting area was characterized into two kinds according to whether it is smooth or not. For a smooth area, simple algorithm-median filter-to restore the pixels＇ value was used. For a non-smooth area, however, it＇ s hard for the median filter algorithm to restore the edges, for it doesn＇ t consider any geography information of the image. To make up for this, the method of the curvature-driven image inpainting was adopted, which makes use of the geography information of the image, to restore the missing information of the damaged image. For the sequence of the process, the time needed to reach the current damaged point was adopted, which can be calculated by the fast-marching method, to determine the order of sequence to be inpainted. The experiments justify the advantage of the proposed method.

关 键 词：图像修补 中值滤波 扩散 快速推移法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]