Deepfake Detection via Eye State Sequence Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for detecting deepfake videos face challenges in accurately identifying authenticity due to degradation from video compression and the lack of feature extraction for eye state changes, particularly closed eye states, leading to inefficiencies in recognition accuracy.
Innovation Solution
A deepfake video detection system utilizing a long recurrent convolutional neural network (LRCN) with long short-term memory (LSTM) for sequence learning, which extracts eye feature models and predicts eye states through a state quantification module, enhancing recognition accuracy by analyzing time-based eye state changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional static image or dynamic image detection methods are used, then the detection process is simple, but the recognition accuracy deteriorates due to video compression degradation and inability to detect eye state changes
Solution Approach 1:
The detection system segments the video into individual frames and further segments each frame to extract eye regions specifically. This segmentation allows the system to focus computational resources on the most discriminative features (eye states) rather than processing the entire video frame, thereby improving recognition accuracy while managing system complexity
Solution Approach 2:
The patent transitions from static 2D image analysis to temporal sequence analysis by processing video frames as time-series data. By adding the time dimension and analyzing eye state transitions across multiple frames, the system captures dynamic characteristics that static images cannot detect, significantly improving recognition accuracy
2Measurement precision
If more sophisticated recognition methods are used to approach 100% recognition accuracy, then the recognition accuracy improves, but the processing time and computational resources increase
Solution Approach 1:
The system extracts only the eye regions from video frames, isolating the most discriminative features for deepfake detection. By extracting and focusing solely on eye state changes rather than analyzing entire facial images or full video content, the system achieves high recognition accuracy while reducing computational burden and processing time
Solution Approach 2:
The patent applies partial action by focusing detection efforts only on the eye region rather than the entire face or video. This selective approach uses sufficient computational resources to achieve high accuracy on the critical eye state analysis without the excessive processing time that would result from analyzing all video content in detail
3Productivity
If conventional image detection methods are used, then the processing speed is fast, but the ability to detect flaws between frames deteriorates due to lack of time continuity analysis
Solution Approach 1:
The system maintains continuity of useful action by analyzing eye state changes across continuous video frames rather than treating each frame independently. This temporal continuity analysis allows the system to detect subtle flaws and inconsistencies in eye behavior that span multiple frames, improving detection reliability while maintaining efficient processing through focused feature extraction
Data Source
AI summary
A deepfake video detection system, including an input data detection module of a video recognition unit for setting a target video; a data pre-processing unit for detecting eye features from the face in the target video; a feature extraction module for extracting eye features and inputting the eye features to a long-term recurrent convolutional neural network (LRCN); and then using a sequence of long-term and short-term memory (LSTM) of a learning module; performing sequence learning; using a state prediction module to predict the output of each neuron, and then using a long and short-term memory model to output the quantized eye state, then connecting to a state quantification module, and comparing the original stored data from the normal video and the quantified eye state information of the target video, and outputting the recognition result by an output data recognition module.