Image Falsification Detection via Color Conversion Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image falsification detection methods struggle to differentiate between normal color conversion and intentional image editing, leading to false positives or requiring high learning costs, and lack a robust mechanism to handle color conversions outside predefined models.
Innovation Solution
An information processing device and method that estimates color conversion applied to an edited image, aligns and compares it with a reference image, using a high-degree-of-freedom model to determine if the editing is normal color conversion or falsification, employing color correspondence relationships and interpolation to enhance detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a low-degree-of-freedom color mapping method (gain, offset, gamma) is used, then the calculation is simple, but color conversion outside this model cannot be detected and is incorrectly classified as falsification
Solution Approach 1:
The patent changes the parameters of the color mapping method from a limited set (gain, offset, gamma) to a more comprehensive set including LUT (Look-Up Table) and polynomial expressions. This allows the system to accurately represent various color conversion methods while maintaining computational efficiency through the structured approach of these parameter sets.
Solution Approach 2:
The patent introduces a dynamic determination mechanism that selects the appropriate color mapping method (LUT, polynomial, or traditional gain/offset/gamma) based on the specific characteristics of the color conversion being analyzed. This dynamic adaptation allows the system to balance between computational simplicity and detection accuracy for different scenarios.
2Measurement precision
If machine learning methods are used for falsification detection, then detection performance can be improved, but large learning costs and computational resources are required
Solution Approach 1:
The patent employs lightweight mathematical models (LUT, polynomial expressions) that can be quickly computed and discarded for each analysis, replacing the need for expensive, resource-intensive machine learning models. These simpler models provide sufficient detection accuracy without requiring large training datasets or significant computational resources.
Solution Approach 2:
The patent substitutes machine learning-based detection with a deterministic mathematical approach using color mapping analysis. This replacement eliminates the need for complex training processes and large computational resources while maintaining effective falsification detection through direct mathematical comparison of color conversion parameters.
3Device complexity
If conventional color mapping methods are used, then computational complexity is low, but they cannot accurately represent complex color conversions leading to false positives
Solution Approach 1:
The patent creates a composite approach by combining multiple color mapping methods (LUT, polynomial expressions, and traditional gain/offset/gamma) into a unified analysis framework. This composite structure allows the system to leverage the strengths of each method while maintaining computational efficiency, accurately representing complex color conversions without excessive computational complexity.
Data Source
AI summary
Information processing with detection of the falsification of an image is disclosed. In one example, an information processing device includes a color conversion unit that makes an estimate of color conversion applied to a second image generated by image editing on a first image, and outputs a color conversion trial result in which color conversion according to the estimate is applied to the first image. A judgement unit compares the color conversion trial result with the second image to judge the presence or absence of falsification in the second image. The technology can be applied, for example, to a falsification detection system that detects whether image editing for falsifying an image has been performed when the image is uploaded to social media.