采用顺序修复的样本例图像修复算法  

Image Inpainting Algorithm Base on Sequential Repaired Exemplar

作　　者：李旭峰[1] 王静[2] 王国东[1] 刘红敏[2] 王志衡[2] 

机构地区：[1]河南理工大学电气工程与自动化学院,焦作454150 [2]河南理工大学计算机科学与技术学院,焦作454150

出　　处：《计算机系统应用》2017年第9期219-223,共5页Computer Systems & Applications

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)

摘　　要：本文提出了一种采用顺序修复的样本例图像修复算法,该方法在原Criminisi经典图像修复算法的基础上对修复顺序进行新的尝试.原Criminisi经典算法的修复顺序通过计算优先级得出,随着修复的深入优先级逐渐趋近于0,导致算法失去作用.为解决该问题,本文采用顺序修复的方法来代替优先级决定顺序,避免出现算法失去作用的情况;同时本文提出的"倒L"型样本模板来增强结构的传播能力、提高匹配的正确率.实验结果证明,本文的修复算法相对Criminisi算法具有优势并取得很好的修复结果.This paper proposes a novel image inpainting algorithm based on exemplar repaired in a sequential manner.This method makes a new experiment on inpainting order of Criminisi classical algorithm. The inpainting order of the classical algorithm is obtained by calculating the priority. But the priority is tending to 0 with the sustained inpainting. As a result, the algorithm will fail and produce error inpainting results. In order to solve this problem, we replace the priority determined order with a sequential manner, which can keep the algorithm working all the time. Moreover, we propose a novel exemplar pattern like an inverted L. The pattern in our method can help to enhance the ability of structure propagation and improve correction rate of patch matching. Experimental results prove that the novel algorithm can achieve better visual effects.

关 键 词：顺序修复 图像修复 样本块 匹配 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]