Application of cluster validity criteria to Rock-Paper-Scissors game judgment

This paper considers an application of cluster validity indexes to automated judgment in Rock-Paper-Scissors (RPS) game. In the task, the number and the locations of hands in RPS pictures are first estimated by c-Means-type clustering in conjunction with cluster validity measures. Then, the judgment of winner/loser is performed considering the difference among compactness of clusters. After preprocessing of extracting hand region coordinates from RPS pictures, Hard c-Means (HCM) and Fuzzy c-Means (FCM) is applied in conjunction with cluster validity measures for searching for the optimal number and locations of hands, and the applicability of noise rejection is compared in several experiments. The compactness of clusters is also measured by the same concept used in cluster validity measures. The experimental results imply that FCM with noise rejection used with cluster validity measures works better in RPS game judgment without following complex image processing.