A Novel Contour Extraction Approach Based on Q-Learning

Contour extraction is an important and challenging issue in image processing. The common contour extraction approaches are sensitive to the initial searching position and image noise, and the extracted contours are coarse. In this paper, we propose a novel contour extraction approach based on Q-learning. According to the grayscale gradient value and similarity in gray space, the Q-learning agent searches and pursues the optimal contour in a step-by-step manner. In order to accelerate the Q-learning speed, we suggest state-reduction and exploration restriction measures. From the experimental results, the novel contour extraction approach based on Q-learning is effective