Image Registration via Optimal Transport and Epipolar Geometry

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional image registration methods are inadequate for registering multimodal images of 3D scenes, as they often require feature matching which is not achievable when images are of different modalities, leading to insufficient accuracy and complexity in computational procedures.

Innovation Solution

The proposed solution utilizes an optimal transport (OT) method that incorporates epipolar geometry as a regularizer to generate a registration map for image registration, allowing for the registration of multi-modal images without the need for camera calibration.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional image registration methods are used, then feature matching can be performed, but registration is not possible when images are of different modalities

Engineering Contradiction:
Improveregistration capabilityVSAvoidregistration accuracy
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent introduces an optimal transport plan as an intermediary that operates in a modality-agnostic feature space. Instead of directly matching features between different modalities, the method transforms features from both images into a common latent space where transport plans can be computed, thereby enabling registration across modalities without direct feature comparison

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces traditional feature matching mechanisms with an optimal transport framework. Instead of relying on similarity measures between corresponding features, the method formulates registration as a mass transport problem, using entropy-regularized optimal transport to compute correspondence between image regions regardless of modality differences

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

2Measurement precision

If conventional image registration methods are used, then registration can be performed, but computational procedures become complex

Engineering Contradiction:
Improveregistration accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent changes the parameter space by transforming image features into a latent representation where optimal transport can be efficiently computed. By modifying the feature space parameters and using entropy regularization, the method simplifies the computational procedure while maintaining registration accuracy

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent performs preliminary feature extraction and transformation into a common latent space before applying optimal transport. This preliminary action prepares the data in a form that enables efficient computation of registration plans, avoiding complex iterative matching procedures

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4384982B1Systems and methods for image processing based on optimal transport and epipolar geometry
Publication Date: 2025.06.18 MITSUBISHI ELECTRIC CORP
  • EP4384982B1 patent drawingFigure 1A
  • EP4384982B1 patent drawingFigure 1B
  • EP4384982B1 patent drawingFigure 1C

AI summary

Systems and methods for image processing for determining a registration map between a first image of a scene with a second image of the scene, include solving an optimal transport (OT) problem to produce the registration map by optimizing a cost function that determines a minimum of a ground cost distance between the first and the second images modified with an epipolar geometry based regularizer including a distance that quantifies the violation of an epipolar geometry constraint between corresponding points defined by the registration map. The ground cost compares a ground cost distance of features extracted within the first image with a ground cost distance of features extracted from the second image.