采用多特征共生矩阵的模板匹配  

Template matching with multi-feature co-occurrence matrix

作　　者：江苏蓬 向伟[1,2,4,5] 刘云鹏[1,2,4,5] 罗海波[1,2,4,5] JIANG Su-peng;XIANG Wei;LIU Yun-peng;LUO Hai-bo(Shenyang Institute of Automation,Chinese Academy of Sciences,Shenyang 110016,China;Institutes for Robotics and Intelligent Manufacturing,Chinese Academy of Sciences,Shenyang 110169,China;University of Chinese Academy of Sciences,Beijing 100049,China;Key Laboratory of Opto-Electronic Information Processing,Chinese Academy of Sciences,Shenyang 110016,China;The Key Lab of Image Understanding and Computer Vision,Shenyang 110016,China)

机构地区：[1]中国科学院沈阳自动化研究所,辽宁沈阳110016 [2]中国科学院机器人与智能制造创新研究院,辽宁沈阳110169 [3]中国科学院大学,北京100049 [4]中国科学院光电信息处理重点实验室,辽宁沈阳110016 [5]辽宁省图像理解与视觉计算重点实验室,辽宁沈阳110016

出　　处：《光学精密工程》2021年第6期1459-1467,共9页Optics and Precision Engineering

基　　金：中科院国防科技创新基金项目(No.Y8K4160401)。

摘　　要：为解决在模板匹配过程中,目标图像发生局部遮挡、背景变化、光照变化以及剧烈非刚性形变等情况而出现的匹配失败问题,本文提出了一种基于多特征融合的共生矩阵模板匹配算法。首先,采用多特征融合的方法提取图像信息。分别提取图像的颜色特征、深度特征、方向梯度直方图(HOG)特征,通过主成分分析(PCA)与K均值聚类的方法实现多通道多特征融合;随后,以共生矩阵作为相似性度量方法,通过统计相似性特征信息来代替直接施加距离计算;最后,计算滑动窗口中每一组像素点对的共生概率,并加权求和作为匹配得分,由此在目标图像上寻找最佳匹配区域。通过实验对比,本文算法的AUC(AreaUnderCurve)得分为0.6586,较目前最好的几种模板匹配算法DDIS-D、DDIS-C、BBS算法分别提高了:7.9%,8.1%,20.2%。采用特征融合的方法能够充分利用图像信息,有效提高匹配的准确率;共生矩阵可以捕获图像的纹理相似性,且这种度量方法仅与共生统计有关,与实际像素无关,能够在一定程度上克服复杂场景对匹配结果带来的影响。实验结果表明本文的方法匹配精度更高、鲁棒性更强。Accurately identifying the target object in complex scenes(such as those involving partial occlu⁃sion,cluttered backgrounds,imbalanced illumination,and nonrigid deformations)is one of the difficulties involved in the study of template matching.To overcome the problem of precise matching under these con⁃ditions,we propose a co-occurrence matrix template matching algorithm based on multi-feature fusion.First,we extracted the color feature,depth feature,and HOG feature of the image.Thereafter,PCA and K-means clustering were used to realize multi-feature fusion.This method employs the co-occurrence ma⁃trix as its similarity measurement;this is different from traditional methods that directly use distance calcula⁃tions.Finally,the matching score was calculated as the weighted sum of the probability of each window.Moreover,the region with the highest score was regarded as the best matching area.The area under the curve score of our algorithm was 0.6586,which was 7.9%,8.1%,and 20.2%higher than those of DDIS-D,DDIS-C,and BBS,respectively.The experimental results showed that multi-feature fusion is an effec⁃tive approach for exploiting image information.With the aid of the co-occurrence matrix,the method can capture the similarities in image textures.Moreover,this measurement focuses on the co-occurrence statis⁃tics alone,rather than the actual pixels;this overcomes the impact of complex scenes on matching.The pro⁃posed method also achieves higher matching accuracy and better robustness than traditional methods.

关 键 词：图像处理 模板匹配 多特征融合 共生矩阵 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]