Fuzzy Hashing for Image Authentication via Entropy Anchor Points
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image authentication methods are vulnerable to subtle manipulations, as cryptographic hashes are sensitive to even minor changes, allowing attackers to bypass anti-piracy measures by modifying images slightly, necessitating a more secure and manipulation-resistant hashing process.
Innovation Solution
The system generates a compression and manipulation-resistant fuzzy hash by identifying anchor points within an image using random number pairs and calculating entropy values, allowing for matching permutations and deviations from the original image, rather than relying on a one-to-one relationship, and produces a checksum that remains consistent under transformations like rotation or compression.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If cryptographic hashes are used to identify duplicate images, then lookup speed is improved, but the system becomes vulnerable to manipulation attacks
Solution Approach 1:
The patent transforms the hash generation approach by changing parameters from exact pixel matching to entropy-based statistical properties. Instead of using cryptographic hashes that require exact matches, the system calculates entropy values for different regions and uses these statistical parameters to create fuzzy hashes that are resilient to manipulation while maintaining fast lookup capability.
Solution Approach 2:
The patent introduces entropy calculation as an intermediary layer between the image data and the hash function. Rather than directly hashing pixel values, the system first computes entropy for various image regions, then uses these entropy values as the basis for fuzzy hashing. This intermediary transformation decouples the hash from direct manipulation of individual pixels.
2Measurement precision
If exact matching is used to identify original images, then identification accuracy is improved, but false positives increase under compression and manipulation
Solution Approach 1:
The system changes the measurement parameter from exact pixel matching to entropy-based similarity. By calculating entropy values for corresponding regions and comparing these statistical measures rather than raw pixel values, the system maintains identification accuracy while becoming tolerant of compression artifacts and minor manipulations that would otherwise cause false positives.
Solution Approach 2:
Instead of requiring complete exact matching of all image data, the patent uses partial matching based on entropy values from key regions. The system identifies anchor points and compares entropy in their neighborhoods, using this partial information to determine image similarity. This approach reduces false positives by focusing on essential structural characteristics rather than requiring perfect reproduction of all details.
3Reliability
If fuzzy hashing is implemented to resist manipulation, then manipulation resistance is improved, but computational complexity increases
Solution Approach 1:
The patent divides the image into multiple regions and calculates entropy for each segment independently, focusing computational effort on key areas. By identifying anchor points and computing entropy primarily in their neighborhoods rather than processing the entire image uniformly, the system achieves manipulation resistance with reduced computational complexity compared to whole-image fuzzy hashing approaches.
Solution Approach 2:
The system performs preliminary actions by first identifying anchor points and calculating their entropy values before generating the final fuzzy hash. This preliminary processing of key regions provides a compact representation that captures essential image characteristics, reducing the computational burden of subsequent hash generation and comparison operations while maintaining manipulation resistance.
Data Source
AI summary
Systems and a method for compression and manipulation-resistant fuzzy hashing are provided. In one or more aspects, a system includes a network interface to receive an image object from a network, and a processor to process the image object. The processing includes generating pairs of random numbers using a hash of pixel data of the image object as a seed. The processing further includes identifying a number of coordinate pairs, within image pixels of the image object, such that coordinate values of each coordinate pair of the identified coordinate pairs approximately matches one pair of the random numbers. A number of first entropy values associated with first sub-areas corresponding to the identified coordinate pairs are determined. An anchor point within the image pixels is identified that has coordinate values corresponding to a sub-area that is associated with a highest entropy value among the determined first entropy values.