A Reinforcement Learning Framework for Parameter Control in Computer Vision Applications

Graham W. Taylor

Abstract: We propose a framework for solving the parameter selection problem for computer vision applications using reinforcement learning agents. Connectionist-based function approximation is employed to reduce the state space. Automatic determination of fuzzy membership functions is stated as a specific case of the parameter selection problem. Entropy of a fuzzy event is used as a reinforcement. We have carried out experiments to generate brightness membership functions for several images. The results show that the reinforcement learning approach is superior to an existing simulated annealing-based approach.