低信噪比图像中目标检测提取方法

Method for Detecting and Extracting an ObJect in Low SNR Image

在低信噪比图像中检测慢运动小目标,无法利用尺寸、形状和纹理特征,使待测目标与噪声干扰形成的虚假目标相区分。基于小目标的慢运动特性及背景噪声的随机性,首先可以用多帧累积的方法检测到目标,再用区域生长法提取目标像素,进而,分析并验证累积帧数目、目标质心移动速度、图像处理算法对单帧检测概率及目标提取效果的影响.通过目标在两种质心运动速度下帧累积效果的对比,指出对非慢运动目标,帧累积法虽然可以有效预测其运动轨迹,但存在较大的单帧漏检率。

When detecting a small slow-moving object in a low SNR multi-frame image, characteristics of size, shape and vein can not be used to distinguish the object from untrue ones formed by noises. Based on the slow-moving characteristics of small objects and the stochastic of background noises, the method of frame integral can be employed first to detect the object, followed by region growing to extract group pixels of it, and further the effects of the number of integral frames, the moving speed of the centroid and the rules of image processing on the single frame detecting probability and extractive results are analyzed and validated. By comparing the effects of one object moving at two different centroid rates, it is demonstrated that the method of frame integral can predict traces of unslow-moving objects, but the opportunity of skipping useful single frame is high.