To extract exactly feature points and match reliably feature point pairs, a robust image registration method based on feature points matching is proposed. Firstly, corners are extracted with the improved Plessey operator which can improve the precision and speed of the feature points extraction, the initial feature point pairs are extracted with the methods of normalized cross correlation (NCC) and bidirectional greatest correlative coefficient (BGCC), and the false feature point pairs are rejected by the random sample consensus (RANSAC) algorithm. Finally, the correct matching feature point pairs are used to realize image registration. The experimental results indicate that this method can fast and accurately extract the corresponding feature points between two images and greatly reduce the probability of false feature points matching. Image registration can be carried out effectively even under varions of conditions of different light, bigger rotation, and repetitive texture.