In this paper, we investigate the problem of dominant point detection on digital curves which consists in nding points with local maximum curvature. Thanks to previous studies of the decomposition of curves into sequence of discrete structures [1, 2, 3], namely maximal blurred segments of width ν [4], an initial algorithm has been proposed in [5] to detect dominant points. However, an heuristic strategy is used to identify the dominant points. We now propose a modied algorithm without heuristics but a simple measure of angle. In addition, two applications are as well presented: (1) polygonal simplication to reduce the number of detected dominant points by associating a weight to each of them, and (2) classication using the polygon issued from the reduced dominant points. The experimental results demonstrate the eciency and robustness of the proposed method.