An Improved SLIC Super-pixel Extraction Algorithm Based on MMTD

Abstract

Super-pixel can improve the computational efficiency of image segmentation algorithm, which is a description of image with more visual significance. The research on super-pixel extraction algorithm has always been a hot spot of current research. SLIC is a common super-pixel extraction algorithm. This method can generate compact and nearly uniform super-pixels. It has a high comprehensive evaluation in the aspects of computing speed, object contour preservation and super-pixel shape. As a result it has been widely used. However, when extracting SLIC (Simple Linear Iterative Clustering) super-pixel blocks, the determination of weight m needs to be specified manually, and the same weight value for different images leads to the problem of unsatisfactory segmentation effect. In order to solve this problem, this paper introduces the medium mathematics used to deal with the fuzzy phenomenon into the super-pixel extraction, and proposes a super-pixel extraction algorithm based on MMTD (Measuring of Medium Truth Scale) and applies it to image segmentation. Firstly, the similarity between Lab distance and coordinate distance is obtained based on MMTD. Then, the weight m of SLIC super-pixel extraction algorithm is adaptively determined by iteration method. Finally, the isolated points are corrected by connected components. The experimental results show that this algorithm has better super-pixel extraction effect than the SLIC algorithm, which can effectively improve the quality of image segmentation in the later stage.