Agglomerative oversegmentation using dual similarity and entropy rate

•The oversegmentation results of our proposed AOS take the local texture into consideration and maintain the local geometric structures of the target objects in images.•A fast agglomerative algorithm is proposed using the complete linkage.•The gain of the entropy rate of random walks utilized in this study not to be updated, which reduces the computational cost.•A novel dual similarity is proposed to dig more information out of the area with lower intensity contrast in an image.