一种多阶段处理的红外小目标检测方法  

A Multistage Process Method for Detecting Small Targets in Infrared Image Sequences

作　　者：姚迅[1] 李德华[1] 孙贤斌[1] 黄飞[1] 

机构地区：[1]华中科技大学图像识别与人工智能研究所,武汉430074

出　　处：《武汉理工大学学报（交通科学与工程版）》2008年第6期1141-1144,共4页Journal of Wuhan University of Technology（Transportation Science & Engineering）

基　　金：教育部重点实验室资助项目(批准号:TKLJ0207)

摘　　要：提出了一种二阶段红外图像序列小目标检测方法.该方法由2个相序的处理阶段组成,即目标的预侦测阶段和目标的确认阶段.在处理的第一阶段,通过快速扫描输入图像,根据视觉特征的显著性抽取潜在的目标集合.这些潜在的目标在第二个处理阶段通过双均值滤波方法被进一步地识别,判断出真目标,剔除假目标.统计分析在理论上证实了这种算法结构的高效和合理性.实验结果表明该方法在运算速度上有显著的优势.This paper proposes a method for detection of small targets in infrared image sequences. The proposed method consists of two serial processing stages - the predetection processing stage and clutter rejection processing stage. The predetection algorithm processes the entire image, in a low computational cost manner, to extract the potential targets based on the saliency of visual feature in each segmented region of image. Subsequently, the clutter rejection operator processes those potential targets passed from first stage using double average gray value filter method to discard spurious targets and identify true targets. A statistic analysis is given to justify the validity and efficiency of the method. The computation complexity of the algorithm is analyzed and compared with other methods. Experimental results suggest that the proposed combining computation architecture of predetection and clutter rejection modules exhibits excellent performance for detecting small target in infrared image sequences, especially in process speed aspect.

关 键 词：红外图像 小目标检测 多阶段处理 特征抽取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]