AI Video Resolution Prediction for Fake 4K Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing metadata-based methods for determining video resolution are unreliable in distinguishing genuine 4K videos from fake 4K videos created by upscaling lower resolution videos, and existing MOS-based methods are time-consuming and expensive.
Innovation Solution
A non-reference video-based AI model trained through deep learning on a dataset of video contents with various resolutions is used to predict the original resolution by downscaling video clips, measuring quality differences, and aggregating scores to determine the actual resolution based on human perception.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If metadata-based resolution determination is used, then the resolution can be obtained easily and quickly, but the method cannot distinguish between genuine 4K video and fake 4K video obtained by upscaling
Solution Approach 1:
The video content is divided into multiple video clips of fixed time length, and each clip is processed independently through downscaling and quality measurement. This segmentation allows the system to analyze different portions of the video to determine the original resolution, making the measurement both thorough and efficient.
Solution Approach 2:
The AI model is pre-trained on a large dataset of video contents with various resolutions before actual resolution prediction. This preliminary training enables the model to quickly and accurately determine original resolution without time-consuming manual analysis during actual use, resolving the contradiction between speed and accuracy.
2Measurement precision
If MOS-based method is used to identify fake 4K videos, then the resolution authenticity can be verified, but the process is very time-consuming and expensive
Solution Approach 1:
The manual MOS-based authentication process is replaced with an automated AI model that performs downscaling and quality measurement. This substitution eliminates the need for human observers to watch and rate videos, dramatically reducing authentication time and cost while maintaining or improving accuracy.
Solution Approach 2:
Instead of having humans directly observe and evaluate the original high-resolution video, the system creates downscaled copies at various resolutions and evaluates these copies using the AI model. This copying approach enables automated analysis without requiring time-consuming human review of the original content.
3Measurement precision
If video is downscaled to multiple resolutions for quality measurement, then the original resolution can be accurately predicted, but the computational complexity increases
Solution Approach 1:
The system dynamically adjusts the downscaling process by selecting specific target resolutions (360p, 480p, 720p, 1080p, 1440p, 2160p) based on the predicted original resolution. This dynamic approach optimizes computational resources by focusing measurements on relevant resolution levels rather than uniformly processing all possible resolutions.
Solution Approach 2:
The system performs downscaling to multiple resolutions beyond what a single resolution check would require, but this excessive action enables accurate prediction by comparing quality scores across different resolution levels. The additional computational effort is justified by the significant improvement in prediction accuracy.
Data Source
Figure 1A
Figure 1B
Figure 2A~2C
AI summary
The present disclosure relates to a method for predicting the original resolution of video contents using a non-reference video-based AI model built by training a training dataset composed of video contents with various resolutions through a deep learning method in order to overcome the limitations of existing metadata-based resolution determination methods. A method for predicting an original resolution of video contents of the present disclosure, comprises: operation (a) of dividing the video contents that are resolution prediction targets into video clips of a fixed time length; operation (b) of downscaling the resolution of each of the video clips to a predetermined resolution; operation (c) of measuring the quality of video clips downscaled in operation (b) using a non-reference video-based AI model in order from low to high resolution, and calculating a quality score difference between video clips of two neighboring resolutions in order from low to high resolution; operation (d) of predicting a resolution of each of the video clips based on the quality score difference; and operation (e) of aggregating the resolutions of each of the video clips to predict an original resolution of the video contents. In the above configuration, the resolutions to be downscaled are 360p, 480p, 720p, 1080p, 1440p, and 2160p.