Camera Fingerprint Comparison Using Universal Noise Floor
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for comparing camera fingerprints face challenges in accurately identifying the best alignment due to similarities in neighboring pixel characteristics, leading to difficulties in distinguishing noise floor calculations and visualizing correlation energy values, which complicates the determination of matching fingerprints.
Innovation Solution
A method that calculates a universal noise floor excluding high-value normalized cross-correlation values and uses transformed correlation energy values, along with a visualization tool employing elliptic paraboloids, to efficiently compare camera and query fingerprints by focusing on the highest NCC values and reducing unnecessary calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a universal noise floor is calculated excluding high-value NCC values, then the differentiation between high and low correlation energy values is improved, but the calculation complexity increases
Solution Approach 1:
The NCC values are segmented into two groups: high-value NCC values (excluded from noise floor calculation) and low-value NCC values (included in noise floor calculation). This segmentation allows the noise floor to be calculated only from the lower-value NCC values, improving the differentiation of correlation energy values while managing calculation complexity through selective exclusion of specific data points.
2Measurement precision
If shift-specific noise floors are calculated for each possible shift, then the accuracy of correlation energy values is improved, but the computational time increases significantly
Solution Approach 1:
The patent extracts and excludes the high-value NCC values from the noise floor calculation process. By removing these specific values that would artificially inflate the noise floor, the method achieves accurate correlation energy differentiation without requiring shift-specific noise floor calculations for every possible shift, thereby reducing computational time while maintaining precision.
3Device complexity
If all NCC values are used to calculate the noise floor, then the noise floor calculation is simplified, but the correlation energy values become similar and difficult to distinguish
Solution Approach 1:
The patent applies local quality by treating different NCC values differently in the noise floor calculation. Specifically, high-value NCC values are excluded from the noise floor calculation while low-value NCC values are included. This selective approach ensures that the noise floor accurately represents the background correlation level without being inflated by high-correlation regions, thereby improving the differentiation of correlation energy values.
4Ease of operation
If the best alignment is determined by maximum NCC value, then the alignment identification is simplified, but the result is affected by image composition and lacks universality
Solution Approach 1:
The patent transforms the alignment determination from using raw NCC values to using correlation energy values. This parameter change involves calculating CE = NCC²/noise_floor, where the noise floor is specifically calculated excluding high-value NCC values. This transformation makes the measurement universal and independent of image composition, as the correlation energy normalization compensates for variations in image characteristics while maintaining ease of alignment identification through maximum value detection.
Data Source
AI summary
The present invention discloses a method for comparing a camera fingerprint to a query fingerprint. An estimate of a camera fingerprint is obtained from a set of one or more images, and the query fingerprint is obtained from one or more images. Using these values, normalized cross-correlations values are determined for each possible alignment of the two fingerprints. Prior to calculating the noise floor, a set including the highest normalized cross-correlation values is identified. A universal noise floor is then calculated excluding this set from the noise floor calculation. The universal noise floor is utilized in calculating a correlation energy for each possible shift. The correlation energy values are then examined to determine whether the camera fingerprint and the query fingerprint match. A visualization tool may also be used to compare the camera fingerprint and the query fingerprint and determine whether the camera fingerprint and the query fingerprint match. The visualization tool may provide a plot utilizing surfaces or heat maps. The plots allow for a quick and easy analysis and determination as to whether the camera fingerprint and the query fingerprint match.