Real-time interactive image segmentation using improved superpixels

Abstract

Since it is very hard to use the conventional segmentation methods to extract an object with a variety of colors, interactive segmentation plays a more and more important role in image processing and is widely used in electronic products. However, performing interactive segmentation pixel by pixel always requires a lot of time. To improve the efficiency, in this paper, superpixels are applied instead of the original pixels for interactive segmentation. Moreover, since the accuracy may be reduced if the superpixel is applied, we also propose several ways to refine the superpixel. We use the variance to decide whether the superpixel should be refined. If so, the superpixel is re-segmented according to the color values of the adjacent superpixels. With the refinement, the edges of superpixels will highly match the boundaries of objects. Simulations show that using the refined superpixel instead of the original superpixel can achieve much better performance for interactive segmentation. Moreover, since the proposed algorithm is superpixel-based and the superpixels can be determined when users are drawing the lines, the computation time is much less than that of other interactive segmentation methods.