Interactive Multi-label Image Segmentation With Multi-layer Tumors Automat

Abstract

Interactive segmentation is useful for selecting object of interest in an image and it continues to be a popular topic. It plays an increasingly important role in image processing and has a wide range of applications. However, performing interactive segmentation pixel by pixel is normally time consuming. This paper presents a new method to improve the segmentation efficiency. The proposed method improves the growcut algorithm by utilizing the super-pixel-level tumors automata (TA), since the super-pixels can supply powerful boundary clues to guide segmentation and can be gathered easily by over-segmentation algorithm. The TA has the similar principle as cellular automata. Given a small number of user-tagged super-pixels, the rest of the image can be automatically segmented by TA. Due to the iterative strategy, user can observe that the processing is faster than the growcut. To obtain the best result, both level set and multi-layer TA approaches are applied. Experiments carried out on the VOC challenge segmentation dataset show that the proposed method achieves state-of-the-art performance.