Optimization of the morphological pyramid for detecting defects on the surface of ferrite cores

The paper describes the optimization of the thresholds in the morphological pyramid detecting defects on the surface of ferrite cores. A morphological defect detector generates a gray level image, called a map of defects, which specifies the areas of the defects as well as the measure of defectiveness for each pixel. The defects of ferrite cores vary in a very wide range of sizes and shapes. The morphological pyramid makes possible detection of defects at various image resolutions. The result is a pyramid of gray level maps of defects. However, the gray level maps are quite noisy, and it is necessary to reject all the unimportant blobs from the map while maintaining the most important ones which allow one to accept or reject the core. A crucial step in processing the gray level maps of defects is thresholding, which converts a gray level map of defects into a binary map in which the defects include only those pixels for which the measure of defectiveness exceeds a specified threshold. The thresholds have to be chosen for all levels of the pyramid, and they need readjustment whenever the lighting conditions are changed. The genetic algorithm allows one to automatically optimize the thresholds. The main difficulty with the genetic algorithm is that it needs a definition of the fitness function assigning a numerical value to the chromosomes in which the thresholds are encoded. The paper presents a method for conversion of the pictorial information contained in the maps of defects into the numerical value of the fitness function. Experimental results confirm the usefulness of the approach described. The optimization of the pyramid was developed for the possible use by the industrial manufacturer of ferrite cores