Feature matching accuracy affects the precision of image registration, which is one of difficult key points for image registration based on features. In order to solve feature points one-to-many mismatching problem caused by the ratio of the closest neighbor and second closest neighbor from one direction, a bidirectional matching method of features for image registration of visible and infrared image is put forward. Firstly, for enhancing the similarity of two images, image reverse and histogram equalization are adopted to process infrared image, so that more consistent features of high repetition rate are extracted. Next, SURF features are matched bilaterally by using the ratio of the closest neighbor and second closest neighbor, to ensure the consistency between feature matching and reduce matching error rate, and then RANSAC is applied to match feature again. Through the two matches, it can realize precise features matching. The experiment results show that the proposed method is better than traditional SURF feature unilateral matching algorithm based on the ratio of the closest neighbor and second closest neighbor in the correct matching ratio and registration accuracy, and the validity of the method suggested is proved.