False video detection method and system based on forensic embedding and context embedding

• Main Authors: CAI FENGLIN, ZHU XIJIE, LI YING, ZHANG HANWEN, ZHU QITAO, TOSHITOKI
• Format: Patent
• Language: Chinese; English
• Published: 30.06.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention relates to the technical field of computer vision, and particularly discloses a forensic embedding and context embedding-based forged video detection method and system, which are used for carrying out local feature enhancement on three sensitive areas during video deep forging, guiding a model to enhance subtle artifacts of local features and position potential anomalies, and improving the accuracy of video deep forging. And joint features of space, frequency and time domains are obtained. Meanwhile, the joint features are divided into non-overlapping analysis blocks, and joint feature embedding is extracted by using forensic feature embedding and context feature embedding according to context dependency and spatial dependency of forensic traces on a video scene; and a Transform attention module is used for providing spatial attention information for two subsequent different networks by utilizing joint feature embedding. And finally, two different networks are utilized to realize counterfeit det