Parameters Determination for Ill-Defined Edge Detection Using Particle Swarm Optimization

Abstract

Segmentation process is one of the preprocess steps in object detection. To achieve segmentation, edge detection is a possible choice. However, if there is noise in image, edge might be ill-defined. Our algorithm for ill-defined edge detection is enhanced in this paper. In particular, we utilized the particle swarm optimization (PSO) in finding optimal parameters in the algorithm, rather than using manual settings as in the original work. There are two data sets, i.e., synthetic and carpal bone data sets, used in the experiment. We found that the intersection over union (IOU) on the blind test data set of the synthetic data set is $0.9200\pm 0.0144$. The result on the blind test carpal bone data set is $0.9228\pm 0.0592$. For the carpal bone data set, we compare the result with that from the original algorithm. The result shows that the enhanced method performs better than the original one. However, there is still a problem in misleading edge direction because of gradient and edge map generation.