Interactive Segmentation Based on Initial Segmentation and Region Merging

Abstract

Interactive segmentation is one of the most popular and challenging segmentation methods. For the purpose to accurately extract objects from different images, this paper presents an interactive image segmentation method based on initial segmentation and region merging that the users only need to roughly mark the regions of the object and background by using markers. First, a Super-pixel based initial segmentation is applied to the original image. After that, the original image will be divided into some regions. Second, a novel region feature based on color histogram and Contourlet transform is computed for each region. Third, merge the regions which are similar to marked object region and merge the regions similar to marked background region, until there are only two kinds of regions - object and background. Compared with other algorithms, our experiment shows that the proposed scheme can reliably and rapidly extract the desired object from the complex background.