Facial Recognition Using Diagonal Covariance for Fast Authentication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional facial recognition methods for self-service devices are time-consuming and require substantial computing resources, making them unsuitable for real-time authentication scenarios.
Innovation Solution
A biometric facial recognition system using a probabilistic neural network (PNN) with a simplified diagonal covariance matrix to quickly determine the similarity between captured and pre-stored images, reducing processing time and resource requirements by calculating mean pixel values and pixel variance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional facial recognition methods (e.g., principal component analysis) are used, then facial recognition accuracy is maintained, but processing time increases substantially and computing resources are consumed heavily
Solution Approach 1:
The patent extracts only the essential features needed for facial recognition by using a simplified diagonal covariance matrix that focuses on pixel variance along the diagonal elements only, rather than computing the full covariance matrix. This extraction of key statistical properties (mean pixel value and diagonal covariance elements) maintains recognition accuracy while dramatically reducing computational complexity and processing time
Solution Approach 2:
The patent changes the parameter representation from full covariance matrices to simplified diagonal covariance matrices, and from raw pixel data to statistical features (mean and variance). This parameter transformation reduces the dimensionality of the data while preserving the essential information needed for accurate facial recognition comparison
2Reliability
If conventional facial recognition methods are used, then recognition capability is maintained, but computing resources for storing images and calculation outputs are consumed heavily
Solution Approach 1:
The patent extracts only the necessary statistical features (mean pixel value and diagonal covariance elements) from the full images, storing these compact representations instead of the original high-resolution images and full covariance matrices. This extraction reduces storage requirements while maintaining the essential information needed for reliable facial recognition
Solution Approach 2:
The patent creates simplified copies of the facial data in the form of statistical features (mean and variance values) rather than storing and processing the original complex image data. These simplified copies enable recognition operations with minimal storage requirements while preserving recognition capability
3Speed
If simplified methods are used to reduce processing time, then speed improves, but facial recognition accuracy may deteriorate
Solution Approach 1:
The patent transforms the data parameters from raw pixel values to statistical features (mean and variance), and from full covariance matrices to diagonal covariance matrices. This parameter change enables faster computation while preserving the essential statistical properties needed for accurate facial recognition, achieving both speed and precision
Data Source
AI summary
Systems and arrangements for performing biometric facial recognition in order to provide access to a device and/or process one or more events are provided. In some examples, one or more images of a user may be received by an entity and pre-processed to obtain a mean pixel value and variance of each image. These values may be stored in association with the image and/or identifying information associated with the user. Upon receiving a request to access a device, the device may capture an image of the user requesting access. The captured image may be processed similarly to the pre-stored images to determine a mean pixel value and variance. The system may compare the determined mean pixel value and variance for the pre-stored images to the captured image to obtain a similarity score. If the similarity score is at or above a predetermined threshold value, the images may be considered a match.