Triplet DNN for Robust Iris Identification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional iris codes based on human-designed features are inefficient, sensitive to variations, and require segmentation of eye images, which can be prone to errors due to occlusions and lighting conditions.

Innovation Solution

A deep neural network (DNN) with a triplet network architecture is used to learn an embedding that maps high-dimensional eye images to a lower-dimensional embedding space, allowing for robust iris identification and verification by processing images in polar coordinates, including the periocular region, to generate a more robust biometric signature.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional iris codes based on human-designed features are used, then the biometric identification system can be implemented with established methods, but the system becomes sensitive to variations and requires complex segmentation that is prone to errors

Engineering Contradiction:
Improverobustness against variations and occlusionsVSAvoidimage segmentation complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces the mechanical segmentation process with a deep neural network that automatically learns and extracts iris features directly from raw eye images. The DNN performs end-to-end learning, eliminating the need for manual segmentation steps including pupil-iris-sclera separation, coordinate mapping, and wavelet transformation, thereby reducing complexity and improving robustness against segmentation errors

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

Solution Approach 2:

The deep neural network performs self-service by automatically adapting to variations in lighting conditions, occlusions, and image quality through its training process. The network learns robust feature representations that are invariant to these variations, making the system self-correcting and reducing sensitivity to environmental factors without requiring complex preprocessing

Inventive Principle:
Principle #25Self-service

2Measurement precision

If wavelet-based iris codes with 2048 bits are used, then comprehensive iris features can be captured, but the processing becomes computationally intensive and less efficient

Engineering Contradiction:
Improveiris feature extraction accuracyVSAvoidprocessing efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent changes the parameter representation from fixed 2048-bit wavelet coefficients to variable-length feature vectors extracted by the deep neural network. The DNN learns to extract only the most discriminative features, reducing the dimensionality while maintaining or improving identification accuracy, thereby increasing processing efficiency

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The deep neural network extracts only the essential and discriminative iris features directly from the input images, eliminating the need to process all 2048 wavelet coefficients. This selective extraction of critical features reduces computational load while preserving measurement precision for biometric identification

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP3485425B1Deep neural network for iris identification
Publication Date: 2023.08.23 MAGIC LEAP INC

AI summary

Systems and methods for iris authentication are disclosed. In one aspect, a deep neural network (DNN) with a triplet network architecture can be trained to learn an embedding (e.g., another DNN) that maps from the higher dimensional eye image space to a lower dimensional embedding space. The DNN can be trained with segmented iris images or images of the periocular region of the eye (including the eye and portions around the eye such as eyelids, eyebrows, eyelashes, and skin surrounding the eye). With the triplet network architecture, an embedding space representation (ESR) of a person's eye image can be closer to the ESRs of the person's other eye images than it is to the ESR of another person's eye image. In another aspect, to authenticate a user as an authorized user, an ESR of the user's eye image can be sufficiently close to an ESR of the authorized user's eye image.