Detecting Boundaries in Natural Images

work with Dave Martin, Xiaofeng Ren, and Jitendra Malik

Consider the image patches at right. Though they lack global context, it is clear which contain boundaries and which do not. The goal of this work is to use features extracted from the image patch to estimate the posterior probability of a boundary passing through the center point. Such a local boundary model is integral to higher-level segmentation algorithms, whether based on grouping pixels into regions or grouping edge fragments into contours.

We formulate features that respond to characteristic changes in brightness and texture associated with natural boundaries. In order to combine the information from these features in an optimal way, a classifier is trained using human labeled images as ground truth. We present precision-recall curves showing that the resulting detector outperforms existing approaches.

Using this local boundary detector as a baseline, we have examined techniques for incorporating more image context. We developed a probabilistic model of curvilinear continuity (as well as junction type) and are able to show that curvilinear continuity yields a quantitative improvement in boundary detection for a large variety of natural images. We have also extended this model to include high-level knowledge of specific object classes in order to perform object specific segmentation.


  • X. Ren, C. Fowlkes, J. Malik. "Cue Integration for Figure/Ground Labeling", NIPS, Vancouver, Canada, (Dec. 2005). [pdf]

  • X. Ren, C. Fowlkes, J. Malik. "Scale-Invariant Contour Completion using Conditional Random Fields", ICCV, Beijing, China, (Oct. 2005). [pdf]

  • D. Martin, C. Fowlkes, J. Malik. "Local Boundary Detection in Natural Images: Matching Human and Machine Performance", ECVP. Paris, France, (September 2003). [ppt]

  • D. Martin, C. Fowlkes, J. Malik. "Learning to Detect Natural Image Boundaries Using Brightness and Texture", NIPS , Vancouver, (Dec 2002). [ps] [pdf] [ppt]

  • D. Martin, C. Fowlkes, J. Malik. "Learning to Detect Natural Image Boundaries Using Local Brightness, Color and Texture Cues", TPAMI 26 (5) [ps] [pdf]

Example Results: