结合图论Normalized Cut方法的彩色图像分割研究

Color Image Segmentation Based on Normalized Cut

归一化彩色图像分割中高斯核函数的参数选择直接影响到分割效果.为了寻找无参数的权值计算方法,提出将彩色图像中的像素点视为RGB空间的三维向量,使用向量的方向和幅值差异来衡量像素点的相似性,再使用Normalized Cut方法进行分割.实验结果表明,该方法能有效区分不同的颜色信息,计算复杂度不大,兼顾了算法的运算时间及分割效果,对不同的彩色图像均能得到准确的分割结果.

In color image segmentation,the traditional Normalized Cut method is sensitive to different parameters of the Gaussian kernel function. In order to calculate the similarity without parameters, the paper regards color image pixels as three-dimensional vectors in RGB space, uses the difference of the vectors in direction and amplitude to measure the similarity of pixels,and then partitions the graph in line with normalized cut criterion. Experiments show that the method can distinguish different color informations effectively,and help to get precise results in color image segmentation.