Natural Scene Interactive Segmentation Using Edge Detection and Maximal Similarity Region Merging

Abstract

A fundamental problem in image processing is image segmentation. The conventional image segmentation methods, to some extent, all suffer from the problem of inaccurate segmentation. A slightly easier and more approachable problem, interactive segmentation, has also received a lot of attentions over the years. In this paper, we propose a fast and simple interactive image segmentation technique. This segmentation process is conducted in two modules. First, the original image is detected by canny edge detection method. Second, the object ofinterest is segmented by using the region merging based on maximal similarity. In this work, color feature is used to measure the closeness between two regions and accordingly the label of the unmarked region is decided. The proposed method extracts the object from the complex background in the image. The effectiveness of the proposed method is validated by experimental results and compared with other method.