Multi-Layer Neural Network for Robust Face Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face recognition technologies using principal component analysis (PCA) face challenges in recognition speed and reliability, especially under pose changes and facial expressions, and are not robust enough for illumination variations.
Innovation Solution
A deep learning-based method utilizing a neural network with multiple layers to extract low-level, mid-level, and high-level features from input images, allowing for robust feature recognition and verification by processing input images through a series of convolutional and pooling layers, and utilizing pre-trained layers to enhance recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If PCA is used to reduce image information to low-dimensional eigenvector space, then information processing efficiency is improved, but recognition reliability deteriorates under pose changes and facial expressions
Solution Approach 1:
The patent segments the feature extraction process into multiple independent layers (first layer extracts first feature information, second layer extracts second feature information from the first). This segmentation allows each layer to specialize in different aspects of feature extraction, improving both efficiency and reliability without requiring complete information processing at each stage.
Solution Approach 2:
The patent transitions from traditional single-layer PCA to a multi-layer neural network architecture, adding the dimension of depth. By extracting features at multiple levels (first feature information, second feature information), the system captures both local and global characteristics, improving recognition reliability while maintaining processing efficiency through hierarchical abstraction.
2Reliability
If a large amount of image information is processed for accurate face recognition, then recognition accuracy is improved, but recognition speed deteriorates
Solution Approach 1:
The patent segments the comprehensive feature extraction task into multiple specialized layers. The first layer extracts basic feature information, and the second layer extracts higher-level feature information from the first. This segmentation allows the system to process large amounts of information efficiently by distributing the computational load across multiple specialized components rather than processing everything in a single stage.
Solution Approach 2:
The patent performs preliminary feature extraction in the first layer before proceeding to more complex analysis in the second layer. By pre-processing and extracting basic feature information first, the system prepares the data in advance, making subsequent recognition operations faster and more efficient while maintaining high accuracy through the use of comprehensive multi-layer features.
3Reliability
If PCA basis vectors are selected to achieve illumination robustness, then feature robustness is improved, but adaptability to pose changes and expressions deteriorates
Solution Approach 1:
The patent segments the feature extraction into multiple layers with different specialization. The first layer can focus on illumination-invariant features, while the second layer captures pose and expression variations. This segmentation allows the system to maintain robustness to illumination changes while simultaneously adapting to pose and expression variations, resolving the contradiction between robustness and adaptability.
Solution Approach 2:
By adding the vertical dimension of multiple feature layers, the patent enables the system to capture both illumination-robust features (in the first layer) and pose/expression-adaptive features (in the second layer). This multi-dimensional feature space allows simultaneous optimization of both robustness and adaptability, as each layer contributes different aspects of feature information.
Data Source
AI summary
A method of recognizing a feature of an image may include receiving an input image including an object; extracting first feature information using a first layer of a neural network, the first feature information indicating a first feature corresponding to the input image among a plurality of first features; extracting second feature information using a second layer of the neural network, the second feature information indicating a second feature among a plurality of second features, the indicated second feature corresponding to the first feature information; and recognizing an element corresponding to the object based on the first feature information and the second feature information.