Visibly better edge-detection using observed image contrasts

Abstract

For an interactive, computer-aided image segmentation tool to work convincingly, the edges marked by it should, where possible, match those observed by the user on the displayed image (illusory edges excluded). An improvement in edge correspondence can be obtained by calculating observed luminance contrasts. These account for both the non-linearity of the screen and for the veiling effect of room illumination on the displayed image. It is shown how a logarithmic transformation of the photopic luminance data can be used in conjuction with the Canny operator to produce a visually convincing contrast edge map. © 1994.