Multimodal Fusion Decision Module for Identity Verification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional multi-modal verification systems lack accuracy in identifying individuals, especially in unconstrained environments, and fail to effectively associate non-biometric modality inputs, leading to performance degradation under varying conditions such as lighting and image resolution variations.

Innovation Solution

A novel multimodal fusion decision module architecture that computes performance parameters like F-measure and scoring accuracy to weight raw scores, enabling association between different modality inputs and generating an overall score for improved decision-making, capable of handling both biometric and non-biometric data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional multi-modal verification systems use maximum probable value for each input to make decisions, then the decision-making process is simple, but the accuracy of individual identification deteriorates due to ignorance of modality inputs with less probable values

Engineering Contradiction:
Improveaccuracy of individual identificationVSAvoidcomplexity of decision module architecture
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the decision-making process into distinct functional modules: a score generation module that produces raw scores for each modality, a weight generation module that computes confidence weights based on performance parameters, and a score fusion module that combines weighted scores. This segmentation allows the system to consider multiple modality inputs beyond just the maximum probable value while maintaining organized and manageable system complexity.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements preliminary computation of performance parameters (such as F-measure and scoring accuracy) and confidence weights before the final decision-making process. By pre-calculating these weights based on historical performance data, the system prepares the necessary weighting factors in advance, enabling more accurate fusion of multiple modality scores without adding significant real-time computational burden.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If the system processes multiple modality inputs with different performance characteristics, then the accuracy can be improved, but the difficulty of detecting and measuring performance parameters increases

Engineering Contradiction:
Improveaccuracy of recognitionVSAvoiddifficulty of computing performance parameters
Core Design Contradiction:
Measurement precisionVSDifficulty of detecting and measuring

Solution Approach 1:

The patent replaces complex manual or manual-like computation of performance parameters with automated statistical computations. The system automatically calculates performance parameters (F-measure, scoring accuracy) and confidence weights using standardized mathematical formulas applied to the score data, eliminating the need for manual analysis and reducing the difficulty of measuring and processing multiple modality performance characteristics.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Reliability

If existing solutions are used in unconstrained environments with varying conditions, then the implementation is straightforward, but the performance degrades due to lighting variations, distance variations, and image resolution issues

Engineering Contradiction:
Improveperformance in unconstrained environmentsVSAvoidadaptability to varying environmental conditions
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent dynamically adjusts the confidence weights assigned to different modality inputs based on their computed performance parameters. When environmental conditions affect certain modalities (e.g., lighting variations affecting image-based recognition), the system automatically reduces the weight of those degraded inputs and increases reliance on more reliable modalities. This adaptive parameter adjustment maintains system reliability across varying environmental conditions.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent creates a universal decision fusion architecture that can handle multiple types of modality inputs (biometric and non-biometric) with varying performance characteristics. The confidence weight mechanism provides a unified approach to combining diverse input types, making the system adaptable to different environmental conditions and input quality levels without requiring separate processing paths for each modality type.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP2523149B1A method and system for association and decision fusion of multimodal inputs
Publication Date: 2023.01.11 TATA CONSULTANCY SERVICES LTD
  • EP2523149B1 patent drawingFigure 1
  • EP2523149B1 patent drawingFigure 2
  • EP2523149B1 patent drawingFigure 3

AI summary

A computer-based system and method to improve the multimodal fusion output at decision level is disclosed. The method proposes computation of a confidence weighted measure for the individual score values obtained for each modality and fuse these new updated scores to get the final decision. These confidence weights are the performance parameters (measured in terms of F-measure) during the offline training step. The process significantly increases the accuracy of the multimodal system.