摘要
针对Criminisi算法中易发生置信项趋向于0,使优先级计算公式失去作用而影响修复结果的"效果丢失",以及无法在已知区域匹配到合适匹配块而造成修复效果失真的问题,提出一种图像修复改进算法.首先通过置信项减去一个固定常数,使置信项不再趋向于0值,保证优先级公式总是发挥作用;然后通过增加有明显识别特征匹配点的优先级使得以该点为中心的待填充块优先进行匹配,减少匹配时的出错率.实验结果证明,该算法较好地克服了Criminisi算法存在的不足,使修复结果更符合视觉感受.
In Criminisi algorithm, the confidence term is easy to tend to 0, which disorders the inpainting sequence and leads to poor inpainting effect. Besides, it also produces distortion caused by the improper matching patch which always be matched in the known area. This paper proposes a new image inpainting algorithm to solve these problems. First, a constant is introduced to make the confidence term not tend to zero, which will keep the priority formula working. Then, the point priority of obvious identified feature is increased, which is utilized to increase the priority of appropriate patch, and reduces the error matching rate. Experimental results prove that the proposed algorithm can overcome the shortcomings of Criminisi algorithm and achieves better visual effect.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2016年第7期1131-1137,共7页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61401150
61472119
61572173)
河南理工大学博士基金(B2013-039)
河南理工大学创新型科研团队计划(T2014-3)
关键词
特征优先
图像修复
置信项
最优匹配
feature precedence
image inpainting
confidence term
the best match
