基于增强分形特征的人造目标检测

Artificial target detection based on enhanced fractal feature

自然背景和人造目标对于某些分形特征存在着一定的本质差别,这些差别为目标检测提供一套全新的方法。研究表明,对于自然背景中嵌入少量人造目标这类简单情况,采用单一的分形特征就能取得较好的检测效果。对于复杂情况下的检测,本文采用Sarkar和Chaudhuri的差分计盒(DBC)分数维法计算图像的分维数,并用概率松弛法(PRIA)对分维值进行特征增强。仿真实验表明,该方法具有较好的检测效果。

Several fractal parameters have intrinsic differences between natural background and the man-made targets. These differences present a new approach for target detection. The research indicates that it is an effective method for detecting the man-made objects from a simple background by using a single fractal feature. For target detection under complexity environment, we calculate the Differential Box Counting （DBC） of the image according to Sarkar and Chaudhuri＇s approach, and then use the Probabilistic Relaxation Iteration Algorithms （PRIA） to enhance the fractal feature. Experimental results demonstrate that the proposed method is feasible and effective.