Depth counterfeit image detection method and system combined with multi-scale features

Abstract

The invention discloses a deep counterfeit image detection method combined with multi-scale features, and specifically, the method comprises the steps: firstly, expanding a data set based on anti-fact causal reasoning, and generating anti-fact and fact samples, so as to prevent a model from being interfered by pseudo-correlation features, and guarantee that the model learns unbiased feature expression; subsequently, a dual-flow prediction network is constructed that uses sufficiently underlying and higher-level convolutional features to capture different types of counterfeit traces. A multi-scale feature extraction module based on a feature pyramid is utilized in each stream to enrich feature representation, and a classifier is arranged at the tail end of each stream. After multi-feature fusion, a deeper convolution module is utilized to learn higher-level semantic information from previously learned combined features, and then a classifier is used for prediction. And finally, combining the results of the three classifiers arranged at different positions of the network to obtain a final prediction result.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a deep forgery image detection method and system combined with multi-scale features. Background technique [0002] Face forgery represented by deepfake technology has attracted widespread attention in society. Although significant progress has been made in deepfake detection in recent years, there is still a lack of general fake face detectors capable of exposing various traces of forgery. In addition, due to the different resolutions of input videos, different proportions of face regions, and different scales of forged traces, a common processing flow and a standard form of CNN structure (stacking several consecutive convolutional layers followed by a classification module) are used. This leads to the following problems when extracting key discriminative features: [0003] (1) Feature omission. The used model specifically focuses on a certain type of forg...

AI Technical Summary

Problems solved by technology

[0006] In order to solve the above problems, the present invention proposes a deep forgery image detection method that combines multi-scale features, which can solve the problems of feature omission, redundancy and bias, and can further improve the detection accuracy and generalization ability

Images