Automatic extraction of infrared small target based on support vector regression and adaptive region growing algorithm

作者：Liu, RuiMing^*; Yang, Lei; Liu, Erqi; Yang, Jie; Li, Ming; Wang, Fanglin

来源：Optical Engineering, 2007, 46(4): 046402.

DOI：10.1117/1.2724874

• region growing algorithm
• support vector regression
• median filter
• high-pass filter

A region-growing algorithm requires knowledge of the position of seed points in advance and must prespecify a threshold for the homogeneity test. Thus, the process of image segmentation by this algorithm cannot be implemented automatically, which often restricts its application in engineering practice. We propose an adaptive region-growing algorithm for automatically extracting small targets from an IR image. Using support vector regression (SVR), seed points can be detected automatically by regarding the intensity of an image as a surface. Furthermore, an adaptive threshold function is given as the criterion of the homogeneity test instead of a fixed threshold, as used in conventional region-growing algorithms. Our technique can extract small targets without human intervention. We carefully design many experiments to demonstrate its validity in extracting IR small targets.

• 出版日期2007-4
• 单位上海交通大学; 中国航天科工集团公司