Evaluation of edge operators using relative and absolute grading

The second approach, names absolute grading, compares any operator to a manually constructed key or target scene. This key is structured to represent the principal edge or edges in the original scene. Absolte grading scores operators on their ability to detect as much of this key edge and as little of the rest of the scene as possible.

Both these techniques provide feedback information that can be used to improve threshold levels, system parameters, and guide operator improvement. They exhibit great potential as valuable tools in any edge detection application.

Six operators were tested by both grading schemes. They were the Sobel, Robert’s gradient, Roberts square root gradient, gradient, range and Frei-Chen operators. Relative grading revealed the superiority of the Sobel operator in various detection environments. Several operator improvement schemes, including single point elimination, association, and linking demonstrated their ability to enhance operator output. Additionally, absolute grading, coupled with a specially selected image test set indicated overall excellence by Robert’s square root gradient in locating selected key edges.