A new segmentation algorithm which was divided into two steps was proposed for an extended target in complex backgrounds by utilizing the K-means clustering and fractal theory. Firstly, the K-means clustering algorithm was improved by using the rough set theory to determine initial cluster centroids. On the basis of K-means clustering segmentation and region connection, the edges of the target and backgrounds were extracted accurately and intactly. After boundary tracking, the potential target regions were detected according to the characteristics of the extended target. Secondly, by giving the function of a fractal dimension changing with the scale, the natural backgrounds in potential target regions were removed by the fractal scale invariance. Then,the background conglutination was eliminated by a mathematical morphology method. The experimental results indicate that the algorithm can segment the extended target in complex backgrounds correctly and reliably, and the segmented target reserves a good contour.%将K-均值聚类方法与分形理论相结合,提出了一种分两个阶段对扩展目标进行分割的方法.在预分割阶段,运用粗糙集理论求取初始聚类中心,在K-均值聚类分割和区域连通的基础上,检测图像边缘并进行边界跟踪,对于获得的目标和背景团块根据扩展目标特性确定目标潜在区域.在进一步分割阶段,给出图像分维数随尺度变化的函数,利用自适应阈值,根据分形理论的尺度不变性进一步抑制预分割结果中的自然背景,并运用形态学开运算消除背景粘连.实验表明该方法能有效并可靠地实现复杂背景下扩展目标的精确分割,分割出的扩展目标轮廓细节保持良好.
展开▼
