Fowlkes Martin Malik 2007

From neurov.is/on

[edit] Notes

They created a simple algorithm that predicted the figure-ground labeling of contours that were hand-labelled. It was correct 70-75% of the time, depending on the size of the window radius (as a percentage of the length of the contour). If humans had all of the information in the local window, then they did much better than the algorithm. However, if the information was restricted to just the contours, the performance was about the same.

One of the major contributions of this paper is that it indicates that humans can perform the figure-ground labeling pretty well just using local cues (not just contours).

[edit] Download

http://www.journalofvision.org/7/8/2/

[edit] Citation

The citation for this paper is:

Fowlkes CC; Martin DR; Malik J (2007). "Local figure-ground cues are valid for natural images.". Journal of Vision 7(8):2: 1-9. 

Edit Cite:Fowlkes Martin Malik 2007 to modify the citation. For information on citing a paper, see Help:Citing.