DUHO image segmentation based on unseeded region growing on superpixels

In this paper, we present a general-purpose segmentation algorithm that works for a large variety of natural scene color images. This DUHO algorithm contains two main steps. First, we generate K superpixels from a given image. Second, we implement an unseeded region growing algorithm to iteratively group these superpixels into appropriate regions to obtain the final segmentation result. Our proposed method preserves the detailed edges and spatial structure in the image. It has three main advantages compared with other region-growing-based segmentation techniques: (1) reduce computational time and depress noise, (2) works on color images, and (3) improve performance. The effectiveness of the DUHO method will be quantitatively evaluated and compared with other published, state-of-art segmentation methods. Our method produces good results from real datasets and requires substantially less computational time.