基于核空间的Otsu阈值法

Otsu Thresholding Method Based on Kernel Space

为了改善传统Otsu分割方法的分割性能,提出了一种基于核空间函数变换的崭新Otsu阈值化分割方法。该方法首先利用再生核空间的核函数将低维空间的样本映射到高维空间,其样本之间的差异性度量采用基于核函数的距离测度;其次得到了一种基于核空间距离的最小二乘法并采用迭代法来估计样本均值;最后得到了基于核函数距离和一维直方图相结合的最小偏差图像阈值化分割方法,将其简称为核空间Otsu阈值法。实验结果表明,基于核空间的Otsu法相对传统Otsu法有更好的分割性能。

To improve the segmentation performance of traditional Otsu thresholding method, a new Otsu thresholding method based on kernel function transformation in the kernel space is proposed.Firstly,samples in low dimensional Euclidean space are transformed to high dimensional reproducing kernel space by the kernel function,and difference measure between samples is adopted by the distance measure based on kernel function.Then,the least square method based on distance in the kernel space is obtained and the new computation method for the sample mean is proposed by the iterated method.Finally,the new least variation thresholding method in the image segmentation is presented based on combination kernel distance with one-dimensional histogram of gray image,and this segmentation method is called the kernel space Otsu thresholding method for short.Experimental results show that the reproducing Hilbert space based Otsu thresholding method is better than traditional Otsu thresholding method on the segmentation performance.