Region-Based Hierarchical Representation for Object Detection

This paper proposes a region-based hierarchical image representation to be used in the context of object detection. The work uses a bottom-up segmentation algorithm and, specifically, the Binary Partition Tree implementation. The different problems that arise when creating and working with a hierarchical region-based representation are analyzed; namely, (i) the creation of the initial partition, that is, the merging and stopping criteria as well we the way to ensure a sufficient accuracy in the scene representation, (ii) the merging criteria used to build the hierarchical representation and (iii) the use of the hierarchical representation for object detection. For steps (i) and (ii), the proposed approach is assessed and compared with previous ones over a subset of the Corel database using well-established partition-based metrics. For (iii), the usefulness of the final region-based representation for object detection is exemplified in different scenarios.