Abstract

Detecting boundaries between semantically meaningful ob- jects in visual scenes is an important component of many vision algo- rithms. In this paper, we propose a novel method for detecting such boundaries based on a simple underlying principle: pixels belonging to the same ob ject exhibit higher statistical dependencies than pixels be- longing to different ob jects. We show how to derive an affinity measure based on this principle using pointwise mutual information, and we show that this measure is indeed a good predictor of whether or not two pixels reside on the same ob ject. Using this affinity with spectral clustering, we can find ob ject boundaries in the image – achieving state-of-the-art re- sults on the BSDS500 dataset. Our method produces pixel-level accurate boundaries while requiring minimal feature engineering.