Neural Network Feature Extraction for Face Authentication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face authentication systems face challenges in accurately matching face images affected by illumination conditions, facial expressions, and orientation, especially when using RGB images, and require costly dedicated devices for three-dimensional face shape data generation.
Innovation Solution
An image processing apparatus that extracts features from data of different modal types using a neural network, converting and comparing feature amounts to determine object identity without relying on special processing or intermediate three-dimensional face shape data, allowing for accurate authentication across various image types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If three-dimensional face shape data is used for face authentication, then authentication accuracy is improved, but device cost increases due to requirement of dedicated devices
Solution Approach 1:
The patent uses two-dimensional face images as simplified copies instead of requiring expensive three-dimensional face shape data. The neural network is trained to extract authentication features from these two-dimensional images, achieving accurate authentication without dedicated three-dimensional scanning devices.
Solution Approach 2:
The patent replaces expensive dedicated three-dimensional data generation devices with ordinary two-dimensional image capturing devices. This substitution uses readily available, inexpensive imaging technology while maintaining authentication effectiveness through advanced neural network processing.
2Ease of operation
If RGB images are used for face authentication, then ease of operation is improved, but authentication accuracy deteriorates due to effects of illumination conditions, facial expressions, and orientation
Solution Approach 1:
The patent transforms the authentication approach by changing from direct image comparison to neural network-based feature extraction. The neural network learns to extract authentication-relevant features while being invariant to illumination conditions, facial expressions, and orientation, thus maintaining accuracy while using simple RGB images.
Solution Approach 2:
The patent replaces traditional mechanical or algorithmic image processing methods with a neural network-based system. This substitution enables the system to automatically adapt to variations in illumination, expression, and orientation without requiring complex preprocessing or multiple image types.
3Measurement precision
If two-dimensional face data and 2.5-dimensional face data are converted into three-dimensional face shape data, then authentication capability is improved, but device complexity increases due to requirement of dedicated devices
Solution Approach 1:
The patent creates a universal neural network-based feature extraction system that can handle multiple types of face data (two-dimensional images, 2.5-dimensional data with depth information) without requiring separate dedicated devices for each type. The same neural network architecture processes all input types, simplifying the overall system.
Solution Approach 2:
Instead of converting all input data to three-dimensional face shape data using dedicated devices, the patent uses the neural network to extract authentication features directly from the original two-dimensional or 2.5-dimensional data, creating an effective abstraction that avoids the need for expensive three-dimensional reconstruction hardware.
Data Source
AI summary
There is provided with an image processing apparatus. An extraction unit extracts a first feature from first data of a first modal type, the first data including information of a first object that is registered, and extract a second feature from second data of a second modal type that is different from the first modal type, the second data including information of a second object for matching. A determination unit determines whether or not the first object and the second object are identical, based on the first feature and the second feature. The extraction unit is trained to extract the first feature and the second feature to be similar when the first object and the second object are identical.