最大后验耦合互信息的图像精确配准算法  

Accurate image registration algorithm based on maximum posterior coupling mutual information

作　　者：赵玉兰[1] 苑全德[2] 孟祥萍[3] ZHAO Yu-lan YUAN Quan-de MENG Xiang-ping(Department of Network Engineering, Jilin Agricultural Science and Technology University, Jilin 132101,China College of Computer Science and Technology, Harbin University of Technology, Harbin 150001,China School of Electrical Engineering and Informat ion Technology, Changchun Inst itute of Technology, Changchun 130012,China)

机构地区：[1]吉林农业科技学院网络工程系,吉林吉林132101 [2]哈尔滨工业大学计算机科学与技术学院,黑龙江哈尔滨150001 [3]长春工程学院电气与信息工程学院,吉林长春130012

出　　处：《计算机工程与设计》2017年第11期3111-3117,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61075076;61075077);吉林省科技厅基金项目(20150203002SF);吉林省教育厅基金项目(2014317)

摘　　要：为提高图像配准精度和稳定性,降低复杂度,提出一种基于最大后验概率耦合互信息的精确图像配准算法。引入图像最大后验概率(posterior A maximum,MAP),利用MAP抑制待匹配区域特征中的背景干扰因素,根据图像退化过程的差异,引入权重因子,通过特征之间对应关系的最小二乘估计,准确提取图像特征;定义最大互信息(mutual information,MI)准则,建立图像特征之间的对应关系,利用高分辨率图像估计对配准参数进行微调。实验结果表明,与当前常用配准算法相比,所提算法能够较准确地从背景中提取目标特征,完成配准,其具有更高的配准精度与更低的复杂度。To improve the accuracy and stability of image registration and to reduce the operation cost,aprecise image registration algorithm based on maximum posterior probability coupling mutual information was proposed.A maximum posterior（MAP）was introduced into the image,MAP was used to suppress the influence of the background factors,and the feature of the matching target was highlighted.According to the difference of the image degradation process,a weight factor was introduced,and image features were extracted accurately by the least square estimation of the corresponding relationship between the features.The maximum mutual information criterion was defined to establish the correspondence between image features,the registration parameters were fine-tuned by high resolution image estimation.Experimental results show that compared with the current commonly used algorithms,the algorithm can extract the target features from the background accurately,and complete the registration.The algorithm accuracy is improved,and the algorithm complexity is reduced.

关 键 词：图像配准 最大后验概率 最大互信息 权重因子 最小二乘估计 高分辨率图像估计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]