基于颜色聚类的光学伪装效果评估背景选取方法

在基于可见光图像的单个目标伪装效果评估中，当目标处于反差较大的两种背景交界区域附近时，直接采用8邻域背景选取方法，将会导致评估结果出现较大偏差。针对这一问题，提出一种基于颜色聚类的背景选取方法。通过该方法对8邻域图像进行颜色聚类后，将目标近邻区域中不包含的颜色类从8邻域图像中剔除，再减去目标区域后即为选择的背景。实验结果表明：采用4～6个颜色聚类中心时，可以有效去除无关的背景区域，增强单个目标伪装效果评估的客观性和准确性；新方法对不同目标位置的适用性较强。

Background Selection Method of Optical Camouflage Effect Evaluation Based on Color Clustering

In the camouflage effect evaluation of single target based on optical images，a large deviation in the evaluation results may be caused by taking the 8-neighborhood image of the target region as the background region when the target is near the boundary between different background areas with large contrast.To solve this problem，a background selection method based on color clustering is proposed.The color classes that are not included in the target neighborhood are removed from the color clustered 8-neighborhood image by the proposed method，and then the target region is subtracted to get the selected background. Experimental results show that the irrelevant background regions can be effectively removed when the image is clustered by 4-6 color cluster centers，so as to enhance the objectivity and accuracy of camouflage effect evaluation. The proposed method is suitable for different target location relationships.