Face Recognition via Gabor Feature Dimensionality Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face recognition systems face challenges due to the high dimensionality of the sample space combined with a limited number of face samples available for training, known as the 'small sample size' problem, which makes many learning algorithms mathematically intractable or empirically unstable, especially when only one image sample per subject is available.
Innovation Solution
The method employs appearance-based techniques that combine Gabor features, Bayesian learning, and support vector machine (SVM)-based classification, normalizing and transforming face patterns into Gabor feature vectors, projecting them into a lower-dimensional intra-subject subspace, and using a weighted summation to classify the probe image based on gallery images, with a square function applied to each component of the projection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If appearance-based approaches are used to represent face patterns as holistic patterns, then recognition capability is improved, but the high dimensionality of the sample space combined with limited training samples causes mathematical intractability and empirical instability
Solution Approach 1:
The patent transforms the high-dimensional face image data into a lower-dimensional feature space using Gabor wavelet transformations and principal component analysis. This dimensionality reduction maps the original pixel-space representations into a compressed feature space that retains essential facial characteristics while eliminating redundant information, thereby resolving the contradiction between maintaining recognition capability and ensuring algorithm stability with limited samples.
Solution Approach 2:
The patent changes the parameter representation from raw pixel values to Gabor feature descriptors that capture local frequency and orientation information. By transforming the feature parameters and applying regularization techniques, the system maintains discriminative power for recognition while reducing sensitivity to variations in lighting, pose, and expression, thus improving both recognition capability and algorithm stability.
2Quantity of substance
If only one image sample per subject is available, then data collection requirements are reduced, but the small sample size problem becomes even more challenging and makes learning algorithms empirically unstable
Solution Approach 1:
The patent performs preliminary dimensionality reduction and feature extraction using Gabor wavelets and principal component analysis on the single available training image. This preliminary processing creates a robust feature representation that captures essential facial characteristics, enabling the algorithm to generalize effectively from limited data and achieve stable learning performance even with only one sample per subject.
Solution Approach 2:
The patent transforms the single training image into a compressed feature vector using Gabor transformations, changing the parameter space from high-dimensional pixel values to lower-dimensional Gabor descriptors. This parameter transformation extracts discriminative features while reducing noise and redundancy, allowing the algorithm to achieve reliable learning stability with minimal training samples.
Data Source
AI summary
Automatic face recognition. In a first example embodiment, a method for automatic face recognition includes several acts. First, a face pattern and two eye patterns are detected. Then, the face pattern is normalized. Next, the normalized face pattern is transformed into a normalized face feature vector of Gabor feature representations. Then, a difference image vector is calculated. Next, the difference image vector is projected to a lower-dimensional intra-subject subspace extracted from a pre-collected training face database. Then, a square function is applied to each component of the projection. Next, a weighted summation of the squared projection is calculated. Then, the previous four acts are repeated for each normalized gallery image feature vector. Finally, the face pattern in the probe digital image is classified as belonging to the gallery image with the highest calculated weighted summation where the highest calculated weighted summation is above a predefined threshold.