Synthetic Fingerprint Generation for Content Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Content recognition systems face performance issues due to differences between clean and transformed media versions, leading to false positives and negatives when comparing temporal and spectral features, such as fingerprints, for content recognition and advertising exposure measurement.
Innovation Solution
A synthetic accuracy measurement service generates synthetic temporal and spectral features, which are used to measure and improve the performance of content recognition systems by comparing them to reference features without altering the original media, allowing for accurate detection of content exposure without requiring actual media files.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If clean media versions are used for reference fingerprint generation, then measurement precision is improved, but reliability deteriorates when comparing with transformed media versions
Solution Approach 1:
The patent creates synthetic fingerprints that copy the characteristics of transformed media versions. Instead of using actual transformed media files, the system generates synthetic representations that replicate the fingerprint features of transformed content, enabling accurate comparison without requiring the original media files.
Solution Approach 2:
The system transforms reference fingerprints by applying the same transformations used in the content recognition system. This includes modifying temporal and spectral parameters to match the transformed media characteristics, ensuring that comparisons are made between equivalent parameter sets.
2Measurement precision
If actual media files are required for benchmarking, then measurement accuracy is improved, but device complexity and operational ease deteriorate
Solution Approach 1:
The patent replaces actual media files with synthetic fingerprints for benchmarking purposes. This synthetic copy approach maintains measurement accuracy while eliminating the need to store, manage, and process large media files, thereby reducing system complexity.
Solution Approach 2:
The system extracts only the essential fingerprint features from media files and uses these extracted features for benchmarking. This extraction approach removes the unnecessary complexity of handling complete media files while retaining the critical information needed for accurate system evaluation.
3Reliability
If transformed media versions are used for comparison, then reliability is improved, but measurement precision deteriorates due to false positives and negatives
Solution Approach 1:
The patent introduces synthetic fingerprints as an intermediary between reference fingerprints and transformed media versions. This intermediary enables accurate comparison by serving as a bridge that captures the characteristics of transformed media without introducing the errors and inconsistencies present in actual transformed files.
4Measurement precision
If manual media transformation and comparison processes are used, then measurement accuracy is improved, but extent of automation deteriorates
Solution Approach 1:
The system performs automatic transformation of reference fingerprints using the same transformation pipeline as the content recognition system. This self-service approach eliminates manual intervention while maintaining measurement accuracy, as the system automatically applies transformations and generates synthetic fingerprints for comparison.
Solution Approach 2:
The patent performs preliminary transformation of reference fingerprints before comparison. By pre-transforming the reference data using the same processes applied to media files, the system automates the benchmarking process while ensuring that comparisons are made between equivalent representations, maintaining accuracy without manual intervention.
Data Source
AI summary
Techniques for synthetic accuracy measurement of a content recognition system are described. According to some examples, a computer-implemented method includes generating, by a provider network, a reference fingerprint for a secondary content (e.g., advertisement) media file; generating, by the provider network, a synthetic fingerprint for a transformed version of the secondary content media file; inserting, by the provider network, the synthetic fingerprint into a stream of fingerprints of a plurality of media files; comparing, by a comparison service of the provider network, the stream of fingerprints including the synthetic fingerprint to the reference fingerprint to generate an indication of a match between the synthetic fingerprint and the reference fingerprint in the stream; and sending the indication of the match to a storage location.