This article introduces a novel region detector based on hierarchies of partitions, so-called Hierarchy-Based Salient Regions (HBSR). This approach enables to combine the clues given by a high quality contour detector with a custom salient region detection procedure. The evaluation of the proposed method HBSR with a standard feature detection assessment framework shows that HBSR outperforms the state-of-the-art methods, in average. These promising results may lead to improvements in many computer vision tasks. © Springer Nature Switzerland AG 2019.