<p>Segmenting an image into meaningful regions remains a challenging computer vision task. Dominant sets clustering, a graph-partitioning algorithm, has been widely used to solve a variety of problems, including image segmentation. This paper focuses on the problem of efficiently constructing a reliable affinity graph gained by integrating two local grouping cues, namely, color and boundary for dominant sets based segmentation. We first over-segment the image into superpixels. The color cue is represented by the mean colors of the inner pixels of superpixels, while the boundary cue is incorporated by calculating geodesic distances over a superpixel graph. The pairwise affinities are then applied into dominant sets clustering algorithm. The experimental results on multiple datasets demonstrate the superiority of our segmentation algorithm in terms of accuracy and efficiency compared with existing popular methods.</p>