Improving graph-based image segmentation using automatic programming

Abstract

This paper investigates how Felzenszwalb’s and Huttenlocher’s graph-based segmentation algorithm can be improved by automatic programming.We show that computers running Automatic Design of Algorithms Through Evolution (ADATE), our system for automatic programming, have induced a new graph-based algorithm that is 12 percent more accurate than the original without affecting the runtime efficiency. The result shows that ADATE is capable of improving an effective image segmentation algorithm and suggests that the system can be used to improve image analysis algorithms in general.