基于轮廓匹配的前视红外建筑目标提取

Building Target Extraction from Forward-looking Infrared Images Based on Contour Matching

针对前视红外图像中建筑物的提取问题，本文利用阈值分割和轮廓匹配，提出了一种基于目标轮廓模板的建筑物自动提取算法。首先，设计了一种目标极性判断方法，它能够自动判断目标极性、选取正确的阈值分割结果，并从分割结果中提取封闭轮廓边缘点；然后，对模板和实时图像的边缘点进行曲线段检测和直线段分割，并在模板的指导下完成实时图像轮廓直线段筛选；最后，通过模板轮廓与筛选后实时图像轮廓的匹配完成建筑目标提取。实验结果表明，该算法实现了目标区域的准确提取，在目标图像尺寸小于120×120时，算法执行时间约为3s，能够满足工程应用的需求。

According to the requirement of building target extraction from forward looking infrared images, a target extrac- tion strategy based on threshold segmentation and contour matching is proposed to realize automatic building target extraction. Firstly, a target polarity judgment method which will judge the target polarity and select the right segmentation result automatically is designed, and the closed contour edge point is extracted from the segmentation resuh. Secondly, curve detection and straight line segmentation are conducted on to the edge points of template and real-time images, and the straight lines are selected under the guidance of target template. Finally, the building target extraction is completed by matching the template contour with the se- lected real-time image contour. Experiment results demonstrate that the proposed method can realize accurate target region extrac- tion and the algorithm execution time is about three seconds when the image size is smaller than 120 × 120, which will satisfy the requirement of engineering application.