摘要：

由于不准确的相似度量无法匹配到合适的样本块,使基于块匹配的图像修复算法在修复建筑场景图像时出现结构和纹理上的不连贯现象。为此,提出利用图像平面结构信息指导块匹配的图像修复方法。通过估计平面投影参数,将已知区域分割成平面,并分析相应的结构信息。根据块偏移和变换的先验概率将平面结构信息转换成目标函数的约束条件,进而指导基于块匹配的图像修复。对颜色项加以改进,引入平面项和梯度项到块距离函数中以度量块间的相似性,减少修复图像中因不准确块匹配而产生的模糊现象。实验结果表明,该算法不仅能够快速有效地对缺损图像进行修复,而且能够保证图像修复后的结构连续性和纹理平滑性。

关键词: 图像修复, 块匹配, 特征提取, 块距离函数, 平面结构信息

Abstract:

As inaccurate similarity measurement fails in matching the appropriate sample patches,image completion algorithms based on patch match tend to cause inconsistent phenomenon on structures and textures when repairing the architectural scene images.In view of this,this paper presents an image completion method using image plane structure information to guide patch match.Begining with estimating plane projection parameters,the known region is segmented into planes,and the translational regularity among these planes is extracted.By means of defining the prior probabilities of the patch offsets and transformations,the plane structure information is converted into soft constraints to guide the ordinary patch match.Integrating the color item,plane item and gradient item into the patch distance function,it measures the similarity of patches,which helps to reduce the image blurring caused by inaccurate patch match.Experimental results suggest that,the proposed method is not only effective in completing the damaged image rapidly,but also able to guarantee the image’s structure continuity and texture smoothness.

Key words: image completion, patch match, feature extraction, patch distance function, planar structure information

中图分类号: 

• TP391