Camera-Specific Generative Models for Image Signature Matching

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image matching technologies face challenges in accurately matching objects across images captured under different imaging conditions, such as varying angles, backgrounds, lighting, and weather, which affects feature distribution and classifier performance in object recognition tasks.

Innovation Solution

A system and method that adapts a universal generative model to create camera-specific models, using Fisher Vectors to encode deviations specific to image content rather than imaging conditions, enabling effective object re-identification across different camera perspectives.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a universal generative model is used for all cameras, then the system complexity is reduced and training is simplified, but the matching accuracy deteriorates due to imaging condition variations

Engineering Contradiction:
Improvesystem complexityVSAvoidmatching accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent segments the universal generative model into camera-specific generative models by adapting the model parameters for each individual camera. This allows each camera to have its own tailored model while maintaining a unified framework, thereby resolving the contradiction between system simplicity and matching accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by customizing the generative model parameters specifically for each camera's imaging characteristics. Instead of using a one-size-fits-all approach, the model is adapted to capture the unique features and conditions of each camera, improving matching accuracy without significantly increasing overall system complexity.

Inventive Principle:
Principle #3Local quality

2Measurement precision

If camera-specific generative models are trained independently, then the matching accuracy improves by accounting for imaging conditions, but the training time and computational resources increase

Engineering Contradiction:
Improvematching accuracyVSAvoidtraining time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary action by pre-adapting the universal generative model parameters for each camera using a dataset captured by that camera before actual matching operations. This preparation step ensures that when matching occurs, the models are already optimized for each camera's specific conditions, reducing the need for repeated training and lowering computational overhead during operational phases.

Inventive Principle:
Principle #10Preliminary action

3Loss of information

If Fisher Vectors encode all feature deviations, then comprehensive image representation is achieved, but the impact of imaging condition variations cannot be distinguished from object differences

Engineering Contradiction:
Improvefeature representation completenessVSAvoidobject re-identification accuracy
Core Design Contradiction:
Loss of informationVSMeasurement precision

Solution Approach 1:

The patent extracts and separates the imaging condition variations from the object features by using camera-specific generative models. These models learn to encode the characteristics of each camera's imaging conditions, allowing the Fisher Vectors to focus on object-specific deviations while filtering out camera-induced variations. This extraction process resolves the contradiction by removing the harmful component (imaging condition noise) while preserving the useful information (object features).

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP3035239B1Adapted vocabularies for matching image signatures with fisher vectors
Publication Date: 2024.04.17 CONDUENT BUSINESS SERVICES LLC
  • EP3035239B1 patent drawingFigure 1
  • EP3035239B1 patent drawingFigure 2A
  • EP3035239B1 patent drawingFigure 2B

AI summary

A method includes adapting the universal generative model of local descriptors to a first camera to obtain a first camera-dependent generative model. The same universal generative model is also adapted to a second camera to obtain a second camera-dependent generative model. From a first image captured by the first camera, a first image-level descriptor is extracted, using the first camera-dependent generative model. From a second image captured by the second camera, a second image-level descriptor is extracted using the second camera-dependent generative model. A similarity is computed between the first image-level descriptor and the second image-level descriptor. Information is output, based on the computed similarity. The adaptation allows differences between the image-level descriptors to be shifted towards deviations in image content, rather than the imaging conditions.