Picture tampering detection method and system based on deep learning

The invention particularly relates to a picture tampering detection method based on deep learning, and the method comprises the following steps: A, inputting a to-be-detected tampered picture into a rough estimation network, and obtaining a strip detection picture close to a tampering edge; b, superposing the tampered graph to be detected and the strip detection graph, and inputting the superposed graph and the strip detection graph into an accurate estimation network to obtain an accurate tampered edge graph; wherein the rough estimation network and the accurate estimation network are pre-trained network models. According to the method, the ambiguity problem that the tampered edge belongs to a tampered part or a non-tampered part is perfectly eliminated by proposing a double-edge method, and a complicated task of directly predicting the tampered edge is converted into two sub-tasks of predicting an area close to the tampered edge and then predicting the accurate tampered edge through a thought from coarse to