基于聚类分析的红外图像配准算法  

Infrared image registration algorithm based on cluster analysis

作　　者：张晓宇 ZHANG Xiaoyu(Xinjiang University of Finance and Economics,Urumqi 830012,China)

机构地区：[1]新疆财经大学,新疆乌鲁木齐830012

出　　处：《现代电子技术》2024年第24期115-119,共5页Modern Electronics Technique

摘　　要：为提高图像处理的精度和效率,并扩展红外图像在实际应用中的使用范围,提出一种基于聚类分析的红外图像配准算法。该算法先构建红外图像观测模型,利用该模型描述理想的红外图像与实际红外图像之间的关系,得到低分辨率的实际红外图像;然后使用最大后验估计方法对低分辨率红外图像进行超分辨率重建;接着以超分辨率重建后的红外图像为基础,利用Harris角点检测算法提取红外图像内角点特征并建立红外图像配准特征集,通过K⁃means聚类算法对红外图像角点特征之间距离、方向特征向量夹角进行计算;最后选择聚类中心,并依据角点特征距离与方向特征向量夹角对红外图像角点特征进行聚类处理,得到红外图像配准结果。实验结果表明:所提算法对低分辨率红外图像的超分辨率重建效果较好,可有效提取红外图像内的角点特征,并实现不同红外图像配准,应用效果较佳。In order to the accuracy and efficiency of image processing,and expand the application range of infrared images in practical applications,a clustering analysis based infrared image registration algorithm is proposed.In this algorithm,an infrared image observation model is constructed,the relationship between ideal infrared image and actual infrared image is described by this model,and the actual infrared image with low resolution is obtained.The maximum a posteriori estimation method is used to reconstruct the low resolution infrared image.Based on the infrared image reconstructed with super resolution,the Harris corner detection algorithm is used to extract the internal corner features of the infrared image and establish the infrared image registration feature set.The distance and direction feature vector angle between corner features in infrared images are calculated by means of the K-means clustering algorithm.The infrared image registration results are obtained by selecting the cluster center and clustering the infrared image corner features according to the angle between the corner feature distance and the direction feature vector.The experimental results show that the proposed algorithm has good effect on super-resolution reconstruction of low resolution infrared images,can effectively extract corner features in infrared images,effectively realize different infrared image registration,and has good application effect.

关 键 词：红外图像 配准算法 聚类分析 观测模型 角点特征 HARRIS角点检测 K⁃means聚类算法 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]