Multi-Scale Deepfake Detection for Compressed Low-Quality Video
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional deepfake detection technologies struggle with low-quality deepfake videos due to image compression, leading to reduced detection performance and vulnerabilities in identifying such attacks.
Innovation Solution
A deepfake detection device utilizing an unsupervised branch-based super-resolution technique and multi-scale deep learning model training method to enhance the resolution of low-quality deepfake videos, enabling effective feature extraction and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If video compression is applied to secure storage space, then storage efficiency is improved, but detection performance deteriorates due to image quality deterioration
Solution Approach 1:
The patent applies preliminary super-resolution enhancement to low-quality deepfake videos before detection. By pre-enhancing the resolution of compressed videos through the super-resolution unit, the system recovers lost high-frequency information and restores features that would otherwise be lost due to compression, thereby maintaining high detection performance while allowing video compression for storage efficiency.
2Measurement precision
If conventional deepfake detection techniques are used, then detection performance is high in benchmark experiments, but detection performance deteriorates drastically for low-quality deepfake videos
Solution Approach 1:
The patent implements a dynamic detection system that adapts to varying video qualities. The multi-scale branch architecture dynamically processes images at different resolutions and scales, allowing the detector to adjust its feature extraction strategy based on the input quality. This enables the system to maintain high detection performance across both high-quality and low-quality deepfake videos, achieving robustness to quality variation.
Solution Approach 2:
The patent introduces multi-scale processing as an additional dimension to the detection pipeline. By analyzing the input image at multiple scales and resolutions simultaneously through parallel branches, the system captures both global and local features effectively. This multi-dimensional approach allows the detector to overcome the limitations of conventional single-scale methods when dealing with low-quality compressed videos.
3Productivity
If image compression is applied, then storage and transmission efficiency is improved, but feature extraction capability deteriorates
Solution Approach 1:
The patent introduces a super-resolution enhancement unit as an intermediary between the compressed video input and the deepfake detector. This intermediary component recovers and restores high-frequency information lost during compression by generating a super-resolved version of the input image. The restored features are then fed to the detector, enabling effective feature extraction without requiring high-quality original videos for transmission.
Data Source
AI summary
A deepfake detection device may include a data input unit that receives an input image including a low-quality deepfake video, a branch-based super-resolution training unit that enhances the resolution of the input image through unsupervised super-resolution training and generates a plurality of super-resolution images having different sizes, and a multi-scale training unit that performs multi-scale training, without resolution conversion, on the plurality of super-resolution images having different sizes, respectively. The multi-scale training unit may synthesize multi-scale training results for the plurality of super-resolution images having different sizes, respectively, and determine whether the input image is a deepfake based on the multi-scale training results.