Object detection using object likelihood and homogeneity likelihood

In this paper, we propose a novel probabilistic framework for detecting object using object likelihood and homogeneity likelihood of segmentations. Our method is based on higher order conditional random fields and uses potentials defined on sets of superpixels (image segmentations) generated using unsupervised segmentation algorithms. These potentials enforce label consistency in image regions and can be seen as a strict generalization of the commonly used pairwise smoothness potentials. The experimental results show that our method improves detection results and obtains better spatial support.