多标号图像分割及其应用

Application of multi-label image segmentation

介绍了一种基于多标号的半自动化图像分割方法。在分割过程中,首先依据高斯权值函数,针对待处理图像建立一个加权图;然后在原始图像中分别标记出属于不同目标区域的像素点;之后,任意选择图像中没有被标号的像素点为作起点,依据所创建的加权图进行随机游走,计算出从当前出发点游走至各个标记像素的概率。通过这种方法,针对图像中未被标号的像素,可以获得一个概率分布图,其中每个概率分布表示未标号像素随机游走到各个标记像素的概率,取概率最大的标记像素作为其所属目标,则可得到一个高质量的分割图像。

A new approach of semi-automated image segmentation based on muhi-label pixel was presented. First, a weighted graph of un-processing image was defined by Gaussian weighting function. Second, the pixels of different goal region were labeled differently in source image. An un-labeled pixel was randomly selected as the starting point. Then, random walking started from this location according to the defined weighted graph, calculating the probability from the start point to each pre-labeled pixels. A probability graph toward each unlabeled pixel was obtained, which represented all probabilities that each unlabeled pixel randomly reached each of pre-labeled pixels. The greatest probability was taken as its objective, therefore, a high-quality image segmentation could be obtained.