Neural networks arbitration for automatic edge detection of 3-dimensional objects

Abstract

The use of Neural Networks for edge detection is in its infancy, and has not as yet been applied in Multiscale analysis. Multiscale edge detection offers a very effective solution to a wide range of feature extraction problems. The work so far reported has focused on region extraction and edge detection of 2-Dimensional objects. Here the noise and illumination effects on the images are less than would be found in the case of a 3-Dimensional object. In the work reported in this paper both the quality of the detected edges and the introduction of the noise and illumination effects due to the third dimension will be considered. This paper reports on investigations into the use of scale space analysis for 3-Dimensional object recognition. The results are then used to form the basis for the use of a Neural Network to carry out Automatic Edge detection, by defining the correct scale at which to apply the Fast Laplacian of the Gaussian operator, during scale space analysis.