Interactive image segmentation with color and texture information by region merging

Interactive image segmentation is able to extract the user-specified foreground objects from the whole image, which remains to be a challenging problem in image processing and computer vision. The traditional pixel-based interactive segmentation is time-consuming and neglects the neighbor information, which is hard to achieve efficient and accurate results. To address this problem, a novel region-based approach is proposed for interactive image segmentation. The algorithm contains three stages: initial segmentation by Simple Linear Iterative Clustering (SLIC) superpixels, region representation combining color features and texture features, region merging based on the region similarity. A new region similarity metric based on the Normalized Cross correlation is raised to guide the region merging process with the aid of user markers. Moreover, since the region merging is a critical step in the entire process, a novel one-stage region merging strategy is exploited to improve the efficiency and robustness of the algorithm. Experiments on various images from the Berkeley dataset is conducted, and the results demonstrate the high speed and effectiveness of the proposed interactive image segmentation method.