基于改进Criminisi算法的病理组织反光图像复原  

Image restoration of reflective area in pathological tissue based on improved Criminisi algorithm

作　　者：何屿彤 赵家龙 张双龙 辜丽川[1] 吴亚文 焦俊[1] HE Yutong;ZHAO Jialong;ZHANG Shuanglong;GU Lichuan;WU Yawen;JIAO Jun(School of Information and Computer Science,Anhui Agricultural University,Hefei Anhui 230036,China;Anhui Fengyang Animal Husbandry and Veterinary Bureau,Fengyang Anhui 233100,China;Anhui Hongsen Networking Company Limited,Bozhou Anhui 236800,China)

机构地区：[1]安徽农业大学信息与计算机学院,安徽合肥230036 [2]安徽省凤阳县畜牧兽医局,安徽凤阳233100 [3]安徽泓森物联网有限公司,安徽亳州236800

出　　处：《阜阳师范学院学报（自然科学版）》2020年第1期69-74,共6页Journal of Fuyang Normal University(Natural Science)

基　　金：国家自然科学基金(31771679);安徽省重点研究与开发计划项目(1804a07020130);安徽省科技重大攻关项目(16030701092);农业农村部农业电子商务重点实验室开放基金(AEC2018010,AEC2018003)资助。

摘　　要：病理组织所含水分导致图像采集时易出现反光现象,影响了图像的质量,降低了计算机对病征的识别效果。为此,本文提出基于Criminisi的改进算法。首先,对病猪器官反光区域进行标记,在对标记图像二值化,对二值化图像用数学形态学闭操作处理,消除空洞,再与未标记的图像叠加,优化修复效果;之后,运用正规化函数定义置信度项,避免出现置信度项迅速衰减;最后,采用加权和的形式定义优先权函数,避免了优先权值为零,导致修复顺序不可靠问题。实验结果表明,改进的Criminisi算法修复顺序更加可靠,填充结果更加自然。The moisture content in the organs of diseased pigs leads to the reflection phenomenon when collecting images,which affects the recognition of symptoms.To improve the reliability of identification,an improved algorithm was proposed based on the Criminisi algorithm.The artificial marker was processed with mathematical morphology closed operation,so that the area to be repaired could be completely covered by the marker,to ensure the correctness of the marker and to improve the repair result.The normalizing function is introduced to define the confidence term to avoid the fast decay of the confidence term.The priority function is defined in the form of weighted sum,which avoids the problem of unreliable repair order caused by zero priority value.The results show that the improved Criminisi algorithm is more reliable and the filling results are more natural.

关 键 词：图像修复 Criminisi算法 数学形态学 反光图像 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]