Iris Recognition Model Training with Near-Infrared Light Correction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Iris recognition systems face challenges in accuracy when used under outside light conditions due to the presence of near-infrared components, leading to mismatches between registered and captured iris codes.

Innovation Solution

An identification model generation apparatus and method that acquire eye images and corresponding image capture status data, infer feature amounts based on these inputs, and perform learning to approximate the feature amounts to ground truth data, generating an identification model that can accurately identify individuals even under outside light conditions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If iris recognition is performed under outside light containing near infrared components, then the system can be used in outdoor environments, but the recognition accuracy deteriorates due to light reflection in the eye

Engineering Contradiction:

Improveoutdoor usabilityVSAvoidiris code matching accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent changes the parameter of light wavelength by using near-infrared light (700nm to 1000nm) instead of visible light for iris capture. This allows the system to operate in outdoor environments while reducing the harmful effects of ambient light reflection, as the near-infrared wavelength is less affected by sunlight and other outdoor light sources.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent replaces traditional iris code comparison methods with a machine learning-based recognition system. The recognition model learns to identify iris patterns by taking into account the specific characteristics of near-infrared light reflection, enabling accurate recognition even when ambient light is present in the capture environment.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Extent of automation

If machine learning is performed using supervisory data combining eye images with scene images and camera angles, then sight-line recognition capability is improved, but the accuracy deteriorates under outside light conditions

Engineering Contradiction:

Improvesight-line recognitionVSAvoidsight-line recognition accuracy

Core Design Contradiction:

Extent of automationVSMeasurement precision

Solution Approach 1:

The patent extracts and removes the scene image and camera angle components from the supervisory data, using only the eye image itself for machine learning. This extraction eliminates the source of error (incorrect sight-line detection under outside light) while preserving the essential iris pattern information needed for accurate recognition.

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS12307819B2Identification model generation apparatus, identification apparatus, identification model generation method, identification method, and storage medium

Publication Date: 2025.05.20 CANON KK

AI summary

An identification model generation apparatus comprises an acquisition unit configured to acquire the image of the eye and data on an image capture status of capturing the image of the eye, an inference unit configured to infer a feature amount of the eye based on the image of the eye and the data on the image capture status, and a learning unit configured to perform learning in the inference unit to approximate the feature amount of the eye obtained from the inference unit to ground truth data that is a ground truth value of the feature amount of the eye corresponding to the image of the eye and the data on the image capture status, and generate the identification model.