Abstract
An image completion method is disclosed. The method includes the following steps: providing an image completion model and establishing a plurality of conditional generative adversarial networks according to a plurality of object types; inputting training images corresponding to the plurality of object types, so as to conduct a corruption training to the plurality of conditional generative adversarial networks; inputting a corrupted image and generating a plurality of repair images based on the different conditional generative adversarial networks; and judging the reasonable possibilities of each repair image and choosing the finished image output by the output interface.