An evaluation of image analysis at benthic sites based on color segmentation Academic Article uri icon


  • Color segmentation is the process of using color to select areas in digital images. The possibility that color segmentation can be implemented to measure percent cover by species at benthic sites is investigated. Color is chosen for image segmentation because some species have a unique color range within any given image. Species of interest are initially identified and matched to a color range by the user. Two methods of color segmentation, designated as automatic and interactive, are compared to manual outline segmentation and a point count method for species discrimination and measurement (primarily of Millepora alcicornis). The accuracy of the color segmentation methods are evaluated by (1) visual examination of species identification accuracy, (2) comparing the histograms of intensity distribution in each color channel for the three selection methods, (3) comparing the average pixel intensities and standard deviations for the three selection methods, and (4) comparing actual final percent area cover values obtained by all four measurement methods. After automatic color segmentation small variations of color within an image lead to an incomplete selection of the species of interest and the selection of areas that do not contain the species of interest. Mean pixel value measurements indicate that automatic color segmentation selects a color range quite different from that selected for a species manually segmented and by interactive color segmentation. Interactive color segmentation employing several color ranges chosen interactively by the user results in accurate and consistent species selection, including areas that might not be easily outlined manually. Percent cover values for M. alcicornis at three representative sites over six years are similar using all four methods of analysis. These results indicate that although there may be conditions where it might be useful, automatic color segmentation based on the algorithms and color ranges selected here cannot be used routinely for accurate species discrimination in images from benthic sites. Interactive color segmentation, on the other hand, works quite well and could be employed routinely.

published proceedings


author list (cited authors)

  • Bernhardt, S. P., & Griffing, L. R.

complete list of authors

  • Bernhardt, SP||Griffing, LR

publication date

  • December 2001