Feature Conversion Learning for Iris Authentication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image authentication systems face challenges when dealing with low-resolution images, particularly in iris authentication, as they require high computational costs to generate high-resolution images using Super Resolution techniques.
Innovation Solution
A feature conversion learning device and method that acquires low-resolution images, reduces them, enlarges them to match the original resolution, extracts and converts features using a learned method based on a loss function that minimizes differences between original and enlarged image features, allowing for accurate authentication without the need for high-resolution images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Super Resolution is used to generate high-resolution images for authentication, then authentication accuracy is improved, but calculation cost increases
Solution Approach 1:
The patent segments the authentication process into two distinct paths: one for high-resolution images (using original extraction) and one for low-resolution images (using enlargement + feature conversion). This allows the system to apply computational resources selectively, avoiding full Super Resolution processing when high resolution is not available, thereby reducing calculation cost while maintaining authentication accuracy where possible
Solution Approach 2:
The patent changes the resolution parameter of the authentication target image by enlarging low-resolution images before feature extraction. Additionally, it transforms the feature space by converting features from enlarged images to match the feature distribution of high-resolution images, enabling accurate authentication without requiring actual high-resolution input, thus avoiding the high calculation cost of Super Resolution
2Use of energy by moving object
If low-resolution images are used for authentication, then calculation cost is reduced, but authentication accuracy deteriorates
Solution Approach 1:
The patent introduces a feature conversion process as an intermediary step between low-resolution feature extraction and authentication comparison. This conversion process adapts the feature distribution of enlarged low-resolution images to match that of high-resolution images, serving as a bridge that enables accurate authentication without requiring actual high-resolution input, thus maintaining accuracy while using low-resolution images
Solution Approach 2:
The patent performs preliminary feature conversion on low-resolution images before authentication, transforming their feature distribution in advance to match high-resolution characteristics. This preliminary adaptation ensures that subsequent authentication comparisons are accurate, effectively preparing the low-resolution data to perform as if it were high-resolution, thereby maintaining accuracy without the computational burden
Data Source
AI summary
A feature conversion learning device is configured to acquire a first image, reduce the first image to a second image having lower resolution than the first image, enlarge the second image to a third image having the same resolution as the first image, extract a first feature that is a feature of the first image and a second feature, convert the second feature into a third feature, and learn a feature conversion method based on a result of comparing the first feature with the third feature.