Face Recognition Feature Extraction via Fixed-Rank Projection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face recognition technologies face challenges in maintaining high recognition rates due to variations in illumination, pose, and environmental conditions, leading to low recognition accuracy compared to other biometric methods.
Innovation Solution
A method involving the generation of a projection matrix with a fixed rank, based on a covariance matrix and a dictionary, to convert high-dimensional input vector data into lower-dimensional feature data, enhancing discriminative features and robustness to noise and environmental changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If face recognition is performed using traditional algorithms, then the recognition process can be completed, but the recognition rate is low due to sensitivity to illumination and pose changes
Solution Approach 1:
The patent transforms the face recognition approach by changing the parameter representation from raw pixel values to discriminative feature vectors. This involves projecting high-dimensional image data into a lower-dimensional space where the features are optimized to maximize inter-class separation and minimize intra-class variation, thereby reducing sensitivity to illumination and pose changes while improving recognition reliability
Solution Approach 2:
The patent applies dimensionality reduction by projecting face images from high-dimensional pixel space into a lower-dimensional feature space using a projection matrix. This dimensional transformation extracts essential discriminative features while eliminating redundant information, enabling robust recognition despite variations in illumination and pose conditions
2Loss of information
If high-dimensional vector data is used for face recognition, then more feature information is preserved, but the computational complexity and processing time increase
Solution Approach 1:
The patent extracts only the most discriminative features from the full high-dimensional face image data by applying a projection matrix that selects and combines specific feature components. This extraction process removes redundant and non-discriminative information, reducing computational complexity while preserving the essential features needed for accurate recognition
Solution Approach 2:
The patent reduces computational complexity by transforming data from high-dimensional pixel space to a compact lower-dimensional feature space. This dimensionality change maintains the critical discriminative information needed for recognition while significantly reducing the computational burden of processing and comparing face images
Data Source
AI summary
A method of converting a vector corresponding to an input image includes receiving input vector data associated with an input image including an object; and converting the received input vector data into feature data based on a projection matrix having a fixed rank, wherein a first dimension of the input vector data is higher than a second dimension of the feature data.