图像失真条件下的模板匹配算法研究  

Research on Characteristics of Template Matching Algorithm Under Image Distortion

作　　者：彭向前 张贻鹏 胡小平[1] 黄泓 PENG Xiangqian;ZHANG Yipeng;Hu Xiaoping;Huang Hong(School of Mechanical Engineering,Hunan University of Science and Technology,Xiangtan Hu’nan 411100,China)

机构地区：[1]湖南科技大学机电工程学院,湖南湘潭411110

出　　处：《传感技术学报》2024年第3期507-512,共6页Chinese Journal of Sensors and Actuators

基　　金：国家自然科学基金项目(61572185);湖南科技大学博士科研启动基金项目(E57110)。

摘　　要：传统模板匹配算法在图像出现局部遮挡、形变等失真情况时适用性不高,为提高图像失真时的匹配性能,提出一种基于等距有效点的局部模板匹配算法。首先,利用局部图像网格处理方法得到候选局部特征向量;然后,根据对应坐标位置的局部相似度提出一种等距有效点的判定方法;最后,通过归一化等距有效点数量及其相似度得到综合相似度。利用局部相似度对失真部分进行筛除,从而使得参与综合相似度计算部分的特征值更准确,进一步增大了综合相似度的真实性和可靠性。所提算法在原始图像集和干扰图像集中的匹配率均为100.0%,在遮挡图像集和形变图像集中则分别为94.1%和83.3%。实验结果表明该算法在图像失真条件下具备较强的抗干扰性,且对噪声和光照变化也具备鲁棒性。Traditional template matching methods suffer from image local occlusion and deformation.To improve the matching performance of template matching algorithm when the image is distorted,a local template matching algorithm based on equidistant effective points is proposed.Firstly,candidate local eigenvectors are obtained by using local image grid processing method.Thereafter,a method to determine equidistant effective points is proposed based on local similarity of corresponding coordinate locations.Finally,the comprehensive similarity is obtained by normalizing the number and similarity of equidistant effective points.Local similarity is used to filter out distorted parts,which makes the eigenvalues of the parts that participate in the calculation of comprehensive similarity more accurate and further enhances the authenticity and reliability of the comprehensive similarity.The matching rate of the algorithm proposed is 100.0% in both the original and the disturbed image set,and it is 94.1% and 83.3% in the occlusion and deformation image set respectively.The experimental results show that by virtue of the proposed algorithm,strong anti-interference is obtained in the condition of image distortion,which is robust to noise and illumination changes.

关 键 词：模板匹配 等距有效点 网格处理 局部匹配 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]