基于区域内一致性和区域间差异性的图像分割

利用区域内方差描述区域内一致性,提出以区域间平均灰度值之差的平方来构建能量函数.运用梯度最陡下降法,推导出能量函数曲线演化方程,并应用于图像分割.研究结果表明:与仅基于区域内一致性或区域间差异性的图像分割算法相比,该算法的图像分割结果更符合实际情况;随着迭代次数增加,基于该算法的图像分割边界变得更平滑;当到达一定的迭代次数时,图像分割结果稳定.

Image segmentation based on intra-region similar and inter-region dissimilar properties

The energy function based on intra-region similar and inter-region dissimilar properties was constructed, and a novel algorithm for image segmentation was presented. The algorithm describes the similar property by intra-region variation and dissimilar property by squares of subtraction of regions mean value. Using gradient-descent methods the energy function is minimized and a curve evolution equation is obtained to segment the image. Compared to the previous method based on either intra-region similar property or inter-region dissimilar property, the experiments show that the results of image segmentation by the developed method tally with the facts. In addition, the segmented image edge is smoothed with increasing iteration number; furthermore, the result of segmentation is not changed until a certain iteration number.