一种适合于目标检测的图像分割方法

An Object Recognition Oriented Image Segmentation Algorithm

在航空图像的目标检测过程中，首先需要把图像中的每个目标都单独地分离开。一般的图像分割方法是把图像分成几大类，同一类目标组成一个个区域。这种分割结果给目标检测造成了很大的困难。本文首先采用最小值滤波对图像进行预处理，尽量地扩大目标之间的间隙。然后用降低分辨率的方法得到初始的分割图像，并用分裂合并法修正分割结果。计算结果显示，分割结果中不同的目标尽管距离很近，也能较好地分离。这个结果对于目标检测有着重要的意义。

One important condition of object detection in aerial image is that different Objects should be separated from each other, The result of general image segmentation algorithm is that the image is divided into several regions, earh of which contains one or more objects of same kind. A new segmentation algorithm is proposed in this paper to meet the needs of object detection for some special kind of aerial images. Firstly, the min-value filtering is employed to separate different objects. Then the image is divided into several parts according to the multi-resolution method. Next, the segmentation hae is modified through splitting and merging. Theoretically analysis and calculating results show tha objects can be separated even in the case they are very close to each other. Such result is very useful for object detecting.