An efficient descriptor and a complete scheme for removing incorrect correspondences are two essentials for accurate feature points matching in image processing and pattern recognition. This paper proposes a new rigid feature points matching algorithm. First, a new feature descriptor based on the local sharpness distribution is proposed to extract features, which leads to a feature set of extremely low dimensionality. With these features, the normalized cross-correlation coefficient and a bidirectional matching strategy are employed to generate the initial feature point correspondences. Then a complete scheme with two rules is proposed to remove the probable incorrect correspondences. Owing to these two rules based on the voting strategy and the ratio of the distances, respectively, the accuracy of feature points matching is remarkably improved. Experimental results show that the proposed algorithm is efficient for feature matching and comparisons with other algorithms show its better comprehensive performance.