Improving foreground segmentations with probabilistic superpixel Markov random fields

We propose a novel post-processing framework to improve foreground segmentations with the use of Probabilistic Superpixel Markov Random Fields. First, we convert a given pixel-based segmentation into a probabilistic superpixel representation. Based on these probabilistic superpixels, a Markov random field exploits structural information and similarities to improve the segmentation. We evaluate our approach on all categories of the Change Detection 2012 dataset. Our approach improves all performance measures simultaneously for eight different basis foreground segmentation algorithms.