Facial Recognition Feature Extraction Using Gabor Filters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition systems face challenges in accuracy and scalability due to factors like gender, age, race, facial expressions, and environmental conditions, and require efficient and robust solutions for on-demand Point-of-Recognition (POR) provisioning.
Innovation Solution
A feature extraction system using multi-orientation and multi-scale Gabor filters to generate Gabor Magnitude Images (GMIs), followed by Local Gabor Binary Pattern (LGBP) maps, which are then used to select regions and generate histograms as feature descriptors, with a double matching routine for robust and verifiable facial recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional face recognition methods are used, then the system is simple to implement, but the recognition accuracy deteriorates under varying conditions such as gender, age, race, facial expressions, and environmental factors
Solution Approach 1:
The face image is divided into multiple local regions around detected feature points (eyes, nose, mouth, etc.). Each region is processed independently to extract local features, which are then combined to form the complete face descriptor. This segmentation approach allows the system to focus on discriminative local patterns while maintaining robustness to global variations in appearance and environment.
Solution Approach 2:
Different Gabor filter parameters (orientation, scale, frequency) are applied to different local regions of the face based on their specific characteristics. The system adapts the filtering parameters to match the local texture and structural properties of each facial region, thereby extracting more discriminative features that are invariant to global variations such as lighting and pose.
2Measurement precision
If exhaustive matching of all stored facial images is performed, then recognition accuracy is maximized, but the processing time increases significantly
Solution Approach 1:
The system performs preliminary filtering by comparing the query face descriptor against a subset of stored descriptors that are pre-selected based on similarity metrics or organizational structures (such as hashing or indexing). This preliminary action eliminates obviously dissimilar candidates before performing the computationally expensive exhaustive matching, thereby reducing the effective search space while maintaining recognition accuracy.
Solution Approach 2:
The system extracts and compares only the most discriminative portions of the face descriptor (corresponding to key facial features) rather than processing the entire descriptor uniformly. By focusing computational resources on the most informative local regions, the system achieves accurate matching with reduced processing time.
3Productivity
If on-demand provisioning is implemented for time-sensitive applications, then the system responds quickly to user needs, but the reliability of recognition deteriorates due to reduced processing time
Solution Approach 1:
The system pre-processes and stores facial descriptors in an optimized format during enrollment, organizing them in advance for rapid retrieval. During authentication, the system leverages this pre-prepared structure to perform quick comparisons without sacrificing the thoroughness of the matching process, thereby achieving both speed and reliability.
Solution Approach 2:
The system replaces traditional sequential mechanical comparison methods with more efficient computational approaches such as vector space modeling, distance metric optimization, or approximate nearest neighbor algorithms. These substitutions enable rapid similarity computation while maintaining the statistical rigor needed for reliable recognition decisions.
Data Source
AI summary
Disclosed is a facial recognition system/method, utilizing modules to perform the following routine: extracting a feature descriptor from a detected feature point of a detected face in an input image frame; and matching the extracted feature descriptor with at least one of a pre-stored facial image that is index-mapped, comprising at least a first and second round of matching, wherein the second round of matching only selects the index-mapped facial images that matched above a pre-defined threshold from the first round of matching. Optionally, the above described steps may be coupled to a Point-of-Recognition (POR) provisioning, enabling an on-demand gate-keeping and/or payment processing for an end-user at an event/venue entry or point-of-sale.