基于分形特征的复杂背景下扩展目标检测

Detection of extended target in complex background based on fractal features

将分形面积度量和分形拟合误差相结合,提出一种复杂背景下扩展目标检测方法。运用分形面积度量进行目标和背景的边缘检测,并结合扩展目标特性确定目标所在区域范围,实现初检。计算原始图像各像素分形拟合误差特征,并运用概率松弛迭代法进行分形特征增强,利用增强特征进一步抑制初检结果中的自然背景。最后运用数学形态学操作剔除背景粘连,实现扩展目标精确检测。实验结果表明:该方法能够有效、可靠地检测复杂背景下的扩展目标,并能较好保持目标的外形轮廓。

Combining fractal area measurement with fractal fitting error, a new algorithm of extended target detection in complex background is proposed. Through analyzing the self-similarity of fractal surface, the fractal feature named area measurement was developed. Using this fractal feature and the characteristic of extended target, potential target regions were detected. Then, the fractal fitting error of every pixel of original image was estimated, and probabilistic relaxation iteration algorithm （PRIA） was utilized to enhance these errors between the natural background and man-made target. Thus, the target was detected by comparing potential target regions and enhanced fractal feature image. Finally, the background conglutinations were eliminated through mathematical morphology operations. The experimental results demonstrate that the algorithm can detect extended target in complex background correctly and the shape details of the target is well reserved.