A proposal on local and adaptive determination of filter scale for edge detection

Abstract

In this work a proposal about local and adaptive scale selection in edge detection is presented. Such a proposal follows Canny's optimality criterion, i.e., the smallest scale that provides a minimum value for the signal to noise ratio should be selected for each edge pixel. The proposal exploits a local regularity, i.e. SNR, measure that is based on a simplified version of the Kitchen-Rosenfeld edge quality evaluator. This measure makes possible a local definition of the scale. The work has been carried out with non-directional gaussian smoothing operators. The integration of the results, obtained at different scales on a pixel basis, into a single edge map is also tackled. Experimental work on real images is presented.

work partially funded by the C. N. R. Targeted Project on Robotics, sub-project URMAD

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.