The present disclosure provides a method for calculating image similarity, including S1: extracting feature points and corresponding feature vectors from a first image and a second image, respectively; S2: pairing all the feature points in the first image with all the feature points in the second image according to similarity by comparing first distances between the feature vectors in the first image and the feature vectors in the second image; S3: sorting the paired feature points according to the similarity from high to low, and selecting top N feature point pairs in the first image and the second image; S4: randomly selecting n reference points from the top N feature point pairs in the first image and the second image, and respectively calculating X-direction and Y-direction relative positions of remaining feature points in the first image or the second image with respect to the reference points; and S5: calculating an X-axis distance according to the X-direction relative positions of the remaining feature points in the first image and the second image with respect to the reference points, calculating a Y-axis distance according to the Y-direction relative positions of the remaining feature points in the first image and the second image with respect to the reference points, calculating the X-axis distance and the Y-axis distance and setting a threshold range to determine whether the first image and the second image are a same image. Detection errors resulting from mismatch of the feature points can be overcome.