城市航空影像中建筑群的FCM聚类分割方法

An Approach of FCM Clustering Segmentation for Building Groups from Urban Aerial Images

在低分辨率城市航空影像中建筑群由于阴影的存在造成其灰度呈现明暗变化 ,采用基于像素级的分割方法以及阈值分割方法均不能得到好的结果。为了充分利用这种明暗变化的信息 ,讨论了一种以图像子块灰度的标准差和直方图的熵作为特征矢量 ,采用基于模糊C -均值 (FCM)的分块聚类方法用于建筑群的粗略分割 ,由于分块有重叠 ,造成边界块的归属不明确 ,因此根据包含边界块的子块的隶属度来确定边界块的归属 ,从而得到了正确的边界区域 ,并利用区域生长和闭合运算对边界进行细化。对实际图像进行实验结果表明 ,该方法是有效的。

The existence of shadows cause changes with light and shade in gray scale in building groups of urban areas from low resolution aerial images. It is not effective on the segment of buildings using the method based on pixels scale and the method based on threshold. In order to take full advantage of this change, a split-block-clustering based on FCM method is discussed for the coarse segmentation of building groups, which considers the variance and the entropy of histogram from image patches as feature vectors .Because the blocks are splitted to overlap each other thus causing ambiguity of border-blocks' ascription, the ascriptions of border-blocks are determined according to grade of membership in image pitches includeing the border-blocks. Thereby the right border regions are attained, and the borders are thinned by region growing and close operation. The result of experiment on actual image show that this method is effective.