Q-Learning Approach for Minutiae Extraction from Fingerprint Image

In this paper we have proposed a Q-learning approach for minutiae extraction from the fingerprint image. Traditional approaches for Minutiae extraction are extremely unreliable in the case of poor quality fingerprint image due to the involvement of image processing steps. This has been improved by using agent based approach SARSA in which agent learns to follow the ridges and stop at the minutiae. One Problem with this approach is that it requires exploring the policy which increases the convergence speed. So, we have proposed a Q-learning approach which is insensitive to the policy of exploration. Agent learns by calculating Q value on the basis of relation between neighbourhood gray scale values of ridges and find original minutiae by selecting maximum Q values. The proposed approach significantly reduces convergence speed due to insensitiveness to the policy exploration.