Deep convolutional neural network information forensics method based on ACBlock

Abstract

The invention provides a deep convolutional neural network information forensics method based on ACBlock, solves the problems that the existing forensics detection method is few in extracted features and low in detection accuracy when facing a video clip with poor quality, introduces an asymmetric convolutional structure ACBlock into a deep convolutional neural network, and improves the forensics detection accuracy. The method comprises the following steps of: training a network by utilizing face contents of an original video set and a tampered video set, taking a result obtained by performing face extraction on a to-be-detected video as a to-be-detected sample, and classifying the to-be-detected sample through a trained convolutional neural network so as to judge whether the frame of image is tampered or not. The ACBlock replaces the original symmetric convolution kernel in a form of combining a plurality of asymmetric convolution kernels, can enhance the feature extraction of the convolution to the central position, overcomes the defect of low detection accuracy due to poor feature extraction capability when the video quality is poor in the prior art, and improves the accuracy of detection and evidence collection.

Description

Technical field[0001]The present invention relates to the technical field of human face video information forensics, and more particularly to a depth convolutional neural network information forensics based on ACBLOCK.Background technique[0002]With the rapid development of information technology, images and videos have become an important information carrier in everyday life, and various digital image processing tools have also been popular, and people can easily tamper with images, make their original, authenticity, and reliable Sexual challenges, especially maliciously changing video, changing video, generated by celebrity or national leaders, has created serious information misleading and security hazards. Therefore, it is necessary to study information forensics, especially for human face video information.[0003]The convolutional neural network has a large number of interest in the image classification field, and the human face video information forensics is used as an image cla...

AI Technical Summary

Problems solved by technology

[0006] In order to solve the problem that the existing face video information forensics detection method faces the poor quality video clips, the extracted features are less and the detection accuracy is not high, the present invention proposes a deep convolutional neural network information based on ACBlock Forensic methods, enhance the ability to extract features, improve detection accuracy

Images