Feature point screening for image matching

Abstract

Generally not all feature points contribute equally to feature matching. Stable, unique and salient feature points are helpful in efficient and accurate image matching. We propose three criteria for screening feature points, which are stability, uniqueness and saliency criterion. Based on stability criterion, robust and stable feature points can be preserved. According to uniqueness criterion, feature points with repeated features would be removed. And points with distinct characteristic information can be retained in accordance with saliency criterion. By eliminating feature points that do not meet these criteria, image matching can be more efficient and accurate. The screening criteria are incorporated into common matching algorithms to verify its effectiveness. Experimental results show that the feature point screening criteria proposed can improve the matching accuracy, repetition rate and matching speed over versatile scenario.