Automatic resolution selection for edge detection using genetic programming

When Genetic Programming is applied to edge detection, the computational cost is generally expensive. When a set of natural images are used to train edge detectors, using their high resolutions is more expensive than using their low resolutions. However, from existing reports, it is hard to find the influence on performance from using different sampling techniques on low resolutions. In this paper, we propose a GP system to automatically select the resolutions of a single training image to train edge detectors. The results of the experiments show that the GP system can effectively evolve edge detectors based on automatic resolution selection. © Springer International Publishing Switzerland 2014.