基于多尺度特征提取的红外图像多目标分割研究  

Research on multi-target segmentation of infrared images based on multi-scale feature extraction

作　　者：陈智 李明 CHEN Zhi;LI Ming(Lanzhou College of Information Science and Technology,Lanzhou 730000,China)

机构地区：[1]兰州信息科技学院,兰州730000

出　　处：《激光杂志》2023年第9期130-134,共5页Laser Journal

基　　金：甘肃省高校创新基金项目(No.2021A-205)。

摘　　要：红外图像由于分辨率过低,导致多目标分割效果较差,为此研究基于多尺度特征提取的红外图像多目标分割方法。通过子带分解多尺度Retinex算法利用增强处理红外图像;利用Linberg尺度空间理论构建增强处理红外图像的多尺度空间,通过二阶自相关矩阵提取红外图像特征点。依据所提取到的特征点,构建红外图像多目标分割聚类的目标函数,采用自适应的进化模糊聚类算法求解红外图像多目标分割聚类问题,实现红外图像多目标分割。实验结果表明,该方法可以有效分割红外图像中的多目标,红外图像分割的信息熵高于5,多目标分割效果较好。Due to the low resolution of infrared image,the effect of multi-target segmentation is poor.Therefore,the multi-target segmentation method of infrared image based on multi-scale feature extraction is studied.Multi-scale Retinex algorithm by subband decomposition utilizes enhanced processing of infrared images;Linberg scale space theory is used to construct a multi-scale space for enhanced processing of infrared images,and the feature points of infrared images are extracted by second-order autocorrelation matrix.According to the extracted feature points,the objective function of infrared image multi-target segmentation and clustering is constructed,and the adaptive evolutionary fuzzy clustering algorithm is used to solve the problem of infrared image multi-target segmentation and clustering,so as to realize infrared image multi-target segmentation.The experimental results show that this method can effectively segment multi-target in infrared image.The information entropy of infrared image segmentation is higher than 5,and the multi-target segmentation effect is better.

关 键 词：多尺度 特征提取 红外图像 多目标分割 子带分解 模糊聚类算法 

分 类 号：TN181[电子电信—物理电子学]