Refined feature fusion method for face counterfeit video detection

Abstract

The invention discloses a refined feature fusion method for face counterfeit video detection and relates to the field of mode recognition. The method comprises steps of carrying out the frame decomposition of a true and false face video, and converting a video format file into a continuous image frame sequence; performing face position detection on the continuous image frame sequence, and adjusting a detection result to enable a face frame to contain a background; cutting a face frame for each frame of image to obtain a face image training set, and training an EfficentNet B0 model; randomly selecting N continuous frames from the face image sequence, and inputting the N continuous frames into an EfficentNet B0 model to obtain a feature map group; and decomposing the feature map group into independent feature maps, re-stacking the feature maps of the same channel according to an original sequence order to obtain a new feature map group, performing secondary feature extraction to obtain afeature vector, connecting the feature vector to a single neuron, and performing final video clip true and false classification by taking sigmoid as an activation function. According to the method, the spatial domain information is reserved, the time domain information is fully extracted, and counterfeiting detection precision is effectively improved.

Description

technical field [0001] The invention relates to the technical field of pattern recognition, in particular to a refined feature fusion method for face forgery video detection. Background technique [0002] In recent years, with the rapid development of network communication technology, the speed and scope of information dissemination have been greatly improved, and hundreds of millions of people can be affected in a short period of time. Therefore, the authenticity of disseminated information is particularly important. False information will disrupt social order and affect people's trust in society. Among them, forged video is a new way of forging information, especially for the forgery of human face. As important information for identity authentication, the face is inherently convenient and unique, and the subconscious of the public will believe more in the information that uses the face as an identity mark. Although there are technical difficulties in forging videos, earl...

AI Technical Summary

Problems solved by technology

Through these technologies, lawbreakers can gain a lot of traffic and attention by disseminating videos to the society, and at the same time earn huge profits, which will cause great reputation loss to individuals and affect the harmonious development of society.

Not only that, exploiting this technology may even endanger national security

Images