Sensor Fingerprint Compression via Random Projection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for managing large databases of digital camera fingerprints are inefficient due to high storage requirements and computational load, and existing compression techniques either lose significant information or increase time complexity.

Innovation Solution

The use of random projection techniques, specifically Johnson-Lindenstrauss projections, to compress sensor fingerprints while preserving distance information, allowing for efficient comparison without decompression and reducing both spatial and temporal complexity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If sensor fingerprints are stored in full resolution (tens of millions of pixels), then identification accuracy is maintained, but storage capacity requirements exceed 10^10 bytes and computational load becomes prohibitive

Engineering Contradiction:
Improvefingerprint identification accuracyVSAvoidstorage capacity
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extracts only the essential identification information from full-resolution sensor fingerprints by computing correlation values between the query fingerprint and database fingerprints. This extraction process converts tens of millions of pixel values into a compact set of correlation metrics, dramatically reducing storage requirements while preserving the ability to identify matching sensors.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent transforms the fingerprint representation from spatial domain (pixel values) to frequency domain (correlation coefficients) through mathematical transformation. This parameter change enables the same identification functionality with significantly reduced data dimensions, as the essential pattern information is preserved in the transformed domain while redundant spatial details are eliminated.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If full-resolution sensor fingerprints are used for database searches, then identification accuracy is maintained, but the computational load per search becomes excessively high

Engineering Contradiction:
Improvefingerprint matching accuracyVSAvoidsearch speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent extracts only the necessary correlation information needed for identification rather than processing complete fingerprint images. By computing correlation coefficients between query and database fingerprints, the system extracts essential pattern matching data while discarding redundant pixel-level information, enabling fast comparison operations.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent implements partial action by computing correlations only for the essential frequency components that carry identification information, rather than processing all pixel data. This selective processing approach maintains sufficient accuracy for sensor identification while dramatically reducing the computational operations required per search query.

Inventive Principle:
Principle #16Partial or excessive action

3Quantity of substance

If sensor fingerprints are compressed using traditional methods, then storage requirements are reduced, but significant information is lost and identification accuracy deteriorates

Engineering Contradiction:
Improvestorage capacityVSAvoidfingerprint identification accuracy
Core Design Contradiction:
Quantity of substanceVSMeasurement precision

Solution Approach 1:

The patent applies parameter changes by transforming fingerprints into the frequency domain through correlation computation. This transformation preserves the essential periodic patterns and structural information that define sensor fingerprints, while naturally compressing the data representation. The correlation-based approach maintains identification accuracy because it captures the fundamental frequency characteristics that are invariant to common image operations.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3314534B1Method and device for searching images
Publication Date: 2021.09.01 POLITECNICO DI TORINO
  • EP3314534B1 patent drawingFigure 1
  • EP3314534B1 patent drawingFigure 2
  • EP3314534B1 patent drawingFigure 3

AI summary

The invention relates to a method and a device (1) for searching for images in memory means (12,13) containing at least one or more images and, for each image, at least one sensor fingerprint associated with said image and related to the sensor that acquired said image, wherein said method comprises a reading phase (PI), wherein a search sensor fingerprint is read, a compression phase (P2), wherein the search sensor fingerprint and at least one of the sensor fingerprints associated with the images are compressed by using a random projection technique, and a searching phase (P3), wherein the images contained in the memory means (12,13) are either selected or discarded by comparing the sensor fingerprints of the images with the search sensor fingerprint.