The present disclosure discloses method and apparatus for image compression. The method includes: acquiring a threshold of a similarity score of an image; acquiring a first quality factor of the image, a first similarity score corresponding to the first quality factor, a second quality factor of the image, and a second similarity score corresponding to the second quality factor; obtaining a functional relationship between quality factor and similarity score of the image by means of curve fitting using the first quality factor, the first similarity score, the second quality factor, and the second similarity score; determining an optimum quality factor of the image according to the functional relationship between the quality factor and the similarity score and the threshold of the similarity score; and compressing the image according to the determined optimum quality factor. With the present disclosure, iterations in image compression can be completed in a very short time.