一种快速的基于CPMC的协同分割算法

A RAPID COSEGMENTATION ALGORITHM BASED ON CPMC

协同分割的一种典型定义是指在给出的一组图像中,将"相似的东西"共同分割出来。提出一种快速的协同分割算法。这种方法首先对一组图像进行特征提取,根据提取的特征确定"相似的东西"在图像中的大致范围,然后使用CPMC(constrained parametric min-cut)算法产生一系列的二值分割结果,最后通过构建随机森林对分割结果与基准之间的"相似度"进行学习,使得分割结果按"相似度"自动进行排列。实验结果表明,所提出的方法在保持良好的分割准确度的同时明显提高了分割的速度。

A typical definition of cosegmentation is to jointly segment ＂something similar＂ in a given set of images. We present a rapid co- segmentation algorithm. The algorithm extracts the features from a set of images first and determines ＂something similar＂ in the approximate range of related images according to the extracted features, then it uses CPMC algorithm to produce a series of binary segmentation results, and finally the ＂similarity＂ between segmentation results and ground truth is learned by creating a random forest, so the segmentation results will be ranked based on ＂similarity＂ automatically. Experimental results show that the proposed method runs high efficiency when segmenta- tion results are satisfactory.