Automatically assigned DDC number: 00637

Manually assigned DDC number: 00637

Number of references: 0

Title: The Statistics of Natural Image Contours



Subject: James H. Elder,Rick Goldberg The Statistics of Natural Image Contours

Description: We report the results of an experimental study in which human subjects are asked to demarcate the contours of a sample of natural images. By analysing the statistics of these contours, we derive a set of empirical probability distributions for three contour grouping cues: proximity, good continuation (curvature), and brightness similarity. We show that of these three cues, proximity is by far the most powerful, followed by good continuation and finally brightness similarity. Using a previously-reported Bayesian contour-grouping algorithm (Elder & Zucker, 1996a), we compare grouping performance based on these empirically-grounded distributions with performance based on the original "guessed" distributions. Surprisingly, although in some cases the distributions differ substantially, we find no qualitative difference in grouping performance. These results suggest that improvements in contour grouping performance will require integration of additional cues beyond local measures of proximit...

Contributor: The Pennsylvania State University CiteSeer Archives

Publisher: unknown

Date: 1998-06-19

Pubyear: unknown

Format: ps



Language: en

Rights: unrestricted


<?xml   version="1.0"   encoding="UTF-8"?>


      <rec   ID="SELF"   Type="SELF"   CiteSeer_Book="SELF"   CiteSeer_Volume="SELF"   Title="The   Statistics   of   Natural   Image   Contours">

            <identifier   Org="ISBN:0792377990"   Paper_ID="SELF"   Extracted="0792377990"   DDC="006.3/7"   Normalized_DDC="00637"   Normalized_Weight="1.0"   />