基于重构的多尺度目标提取与分割

Multi-scale Object Extraction and Segmentation Based on Reconstruction

  • 摘要: 通过形态重构开闭算子构造了一个多尺度的目标提取和分割算法。首先用结构元素nB对影像分别作基于重构的Top-Hat和Bottom-Hat变换,得到所有不能放入nB的亮和暗的目标;然后再用结构元素(n-1)B对影像作开运算,消除所有不能放入该结构元素的目标。那么结果影像中就只剩下能同时放入nB和(n-1)B的目标。对不同尺度的结果影像进行处理,就可以得到不同尺度下的目标分割结果。重构是一个连通区域算子,连通区域是运算的基本单位,所以不会改变影像中边缘的位置,同时不会有新的边缘和虚假的极值出现。

     

    Abstract: A method of multi-scale object extraction and segmentation based on morphological opening and closing by reconstruction is proposed.Top-Hat and Bottom-Hat transformation based on reconstruction with SE(structure element) nB are applied to image to get all objects which can not put into SE nB.Then objects smaller than(n-1)B are removed by opening reconstruction.So the remain objects can only hold SEs smaller than(n-1)B but not nB in the result image.Segmentation of various scale result images can get different scale object's counters.Reconstruction is an operator based on connected region which can exactly preserve the boundaries and edges of image and do not create spurious extrema.

     

/

返回文章
返回