The invention discloses a mobile phone anti-shake function evaluation method based on image fuzziness. The method comprises steps of fixing the position of a mobile phone position, shooting static images, then, starting the vibrating table, shooting a group of images when the anti-shake function of the mobile phone is started and closed; then, calculating image fuzziness based on a deep learning method, carrying out feature extraction by using a deep learning deblurring network, calculating a Wasserstein distance of a feature vector to obtain fuzziness amounts of the two groups of images, andfinally, generating an image fuzziness evaluation result according to the fuzziness amounts. According to the invention, the whole image can participate in ambiguity calculation; compared with an existing evaluation method adopting a local image for calculation, the method has the advantages that the influence of different areas on the calculation result is avoided, the calculation is more reasonable, the result is more accurate, and meanwhile, the calculated ambiguity makes detailed and clear division on the anti-shake function of the mobile phone, so that the objectivity of the evaluation result is improved, and a user can understand and grasp the evaluation result more easily.