针对毫米波图像噪声强、分辨率低的特点,提出了一种新的边缘检测方法。该方法首先根据统计信号处理理论定义了标准化梯度强度这一物理量;然后采用3次B样条函数的一阶导数作为边缘检测算子,由小波变换后得到的图像水平、垂直方向的高频信息,并根据这些信息确定出标准化梯度强度;接着采用单门限的处理得到图像粗边缘;最后对粗边缘施行非最大抑制处理和滤波来得到检测结果。实验证明,该方法可用固定的单门限自动、快速地检测出毫米波图像中人体背景下物体的边缘,满足安检需要。

This paper proposes an edge detection scheme formillimeterwave images, which has low resolution and large noise. It introduces standard gradient strength according to statistical signal processing theory. The basic idea is to employ cubic Bspline as edge detector. Thewavelet transform results reflect the variations for images—along horizontal edges and vertical edges, which are usedwith statistical information together to get the standard gradientmagnitude. For each millimeter wave image a fixed and identical threshold is adopted to detect the image edges roughly. Finally, non2maximum suppression phase and a filter are introduced to get the specific edges. The experiment results demonstrate that the scheme is effective and feasible for detecting the edge of the hidden objects on the background of the human body, which meet the real time requirement in custom.