Fingerprint Generation Using Multi-Level Frequency Sub-band Decomposition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fingerprinting algorithms for information signals, such as those in U.S. Pat. Nos. 8,204,314 and International patent application WO 02/065782, face issues with discriminative power and robustness due to correlation between computed features and susceptibility to noise, leading to inefficient and unreliable identification of content.
Innovation Solution
The method involves decomposing an information signal into multiple frequency sub-bands at varying decomposition levels, calculating spectral properties, comparing these against criteria, and combining results to generate a fingerprint, with differences computed between non-overlapping pairs of sub-bands to reduce redundancy and apply normalization and weight factors to enhance reliability across levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If frequency differences are computed between adjacent frequency sub-bands, then the fingerprint captures spectral characteristics, but the computed frequency differences are correlated which reduces discriminative power
Solution Approach 1:
The patent segments the frequency spectrum into multiple decomposition levels, where each level divides frequency sub-bands into smaller intervals. This hierarchical segmentation allows computing frequency differences at different granularities, reducing correlation between adjacent comparisons while preserving spectral characteristics across multiple scales.
Solution Approach 2:
The patent adds a temporal dimension by computing frequency differences between successive frames at multiple decomposition levels. This multi-dimensional approach (frequency × time × decomposition level) increases discriminative power while the diverse sampling across levels reduces correlation between features.
2Measurement precision
If complex calculations are performed to compute frequency differences and spectral properties, then the fingerprint captures detailed signal characteristics, but the calculation process becomes slow
Solution Approach 1:
The patent performs preliminary computation of spectral properties (such as short-time Fourier transform coefficients) for each frame before computing frequency differences. This preprocessing organizes the spectral data in a structured manner, enabling faster subsequent calculations of frequency differences and comparisons across multiple decomposition levels.
Solution Approach 2:
The patent computes spectral properties and frequency differences at multiple decomposition levels, which may seem excessive, but allows selective use of only the necessary levels for a given application. The hierarchical structure enables computing fingerprints at coarser levels first, providing quick results, with the option to refine at finer levels only when needed.
3Measurement precision
If all mean luminances are compared against each other to create relative ordering, then spatial characteristics are captured, but the process is slow and blocks are correlated which reduces robustness
Solution Approach 1:
The patent segments the spatial domain into blocks and further divides them into sub-blocks at multiple decomposition levels. Instead of comparing all block mean luminances against each other, the method compares luminances within localized groups at each level, reducing the computational complexity from O(n²) to O(n log n) while maintaining spatial characteristic capture through the hierarchical structure.
Solution Approach 2:
The patent adds a decomposition level dimension to spatial comparison, organizing blocks into a hierarchical pyramid structure. This allows comparing spatial characteristics across multiple scales (from fine-grained sub-blocks to coarse-grained blocks), capturing spatial patterns at different resolutions while reducing correlation through the multi-scale approach.
Data Source
AI summary
Method and system for generating a fingerprint representing a portion of an information signal. The method comprises decomposing a portion of the information signal into plural frequency sub bands at a decomposition level, calculating a spectral property of the signal in each of said plural frequency sub bands, comparing each spectral property against a first criterion thereby generating a comparison result, combining each comparison result for constituting the fingerprint, at least once repeating the decomposing, calculating, comparing and combining, wherein for each repetition the decomposing is performed using a decomposition level different from a previous decomposition level.