Abstract
Bottom-up segmentation based only on low-level cues is a notoriously difficult problem. This difficulty has lead to recent top-down segmentation algo- rithms that are based on class-speci fic im age information. Despite the success of top-down algorithms, they often give coarse segmentations that can be signifi- cantly refined using low-level cues. This raises the question of how to combine both top-down and bottom-up cues in a principled manner. In this paper we approach this problem using supervised learning. Given a training set of ground truth segmentations we train a fragment-based segmenta- tion algorithm which takes into account both bottom-up and top-down cues si- multaneously, in contrast to most existing algorithms which train top-down and bottom-up modules separately. We formulate the problem in the framework of Conditional Random Fields (CRF) and derive a novel feature induction algorithm for CRF, which allows us to efficiently search over thousands of candidate frag- ments. Whereas pure top-down algorithms often require hundreds of fragments, our simultaneous learning procedure yields algorithms with a handful of frag- ments that are combined with low-level cues to efficiently compute high quality segmentations.