Reinforcement learning for parameter control of text detection in images from video sequences

This work presents a generalized framework for approaching the problem of optimizing parameters for a multistep computer vision algorithm using reinforcement learning (RL) agent. We then focus on the specific task of text detection in images taken from video sequences for semantic indexing. Employing the fuzzy ARTMAP ANN is an effective way to manage and predict large state and action spaces. Image features in the state space will allow for optimization on a per-image.