• • Input image
    
  • Sobel & Feldman 1968
    
  • Arbeláez et al. 2011
    
  • Dollár & Zitnick 2013
    
  • Our method
    
  • Human labelers
    

• Comparison of our method to several other boundary detection algorithms. Our method detects crisp boundaries between objects while suppressing texture within objects. This is done in an unsupervised fashion and relies only on very simple image features.

• • Download paper
  • 
• Download code

  In Matlab, you can compute crisp boundaries using: >> I = imread('path/to/image');
>> boundary_map = findBoundaries(I); Feel free to checkout our GitHub repository.

Abstract

Detecting boundaries between semantically meaningful objects in visual scenes is an important component of many vision algorithms. In this paper, we propose a novel method for detecting such boundaries based on a simple underlying principle: pixels belonging to the same object exhibit higher statistical dependencies than pixels belonging to different objects. 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 object. Using this affinity with spectral clustering, we can find object boundaries in the image -- achieving state-of-the-art results on the BSDS500 dataset. Our method produces pixel-level accurate boundaries while requiring minimal feature engineering.

Poster

Results

Full results on BSDS 500

Bibtex

@inproceedings{crisp_boundaries,
   author="Phillip Isola and Daniel Zoran and Dilip Krishnan and Edward H. Adelson",
   title="Crisp Boundary Detection Using Pointwise Mutual Information", 
   booktitle="ECCV", 
   year="2014"
}