Face Recognition via Sparse Representation and Regularized Least Squares
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition systems face challenges in uncontrolled environments due to the requirement of large numbers of training samples per class, and they struggle with varying conditions such as lighting and occlusions.
Innovation Solution
A face recognition method combining sparse representation with regularized least squares-based classification, which constructs a new face dictionary using non-zero coefficients from a sparse vector and solves a regularized least squares problem to determine the face identity with minimal class residual.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sparse representation-based classification is used, then face recognition performance is improved, but the system requires large numbers of training samples per class
Solution Approach 1:
The patent extracts only the essential information from training samples by using sparse representation to identify and retain only the most informative samples. The sparse vector a captures the essential features of the test sample, and the new face dictionary T' is constructed using only training samples with non-zero coefficients in a, effectively extracting the most relevant training data without requiring all samples.
Solution Approach 2:
The patent changes the parameter of sample selection from using all training samples to using a subset selected by sparsity criteria. By transforming the problem into a sparse representation problem and using l1 optimization, the system changes how training samples are selected and weighted, allowing better performance with fewer samples.
2Measurement precision
If traditional face recognition methods are used, then system simplicity is maintained, but accuracy in uncontrolled environments deteriorates
Solution Approach 1:
The patent performs preliminary actions by pre-processing training images to create a face dictionary T, and pre-computing sparse representations. The regularized least squares problem is set up in advance with the optimization objective clearly defined. This preliminary preparation allows the system to handle uncontrolled environments effectively while maintaining computational efficiency during actual recognition.
Solution Approach 2:
The patent introduces an intermediary approach by combining sparse representation with regularized least squares. The sparse vector a serves as an intermediary that captures essential features, and the regularized least squares step refines this representation. This intermediary mechanism bridges the gap between simple classification and complex representation learning, achieving high accuracy without excessive complexity.
3Reliability
If more training samples are used, then recognition robustness is improved, but computational complexity increases
Solution Approach 1:
The patent extracts only the necessary training samples by using sparse representation to identify which samples are most informative. The new face dictionary T' contains only training samples with non-zero coefficients in the sparse vector a, reducing the effective number of samples processed while maintaining robustness. This extraction approach preserves recognition reliability without the computational burden of processing all training samples.
4Measurement precision
If sparse representation with l1 optimization is used, then representation accuracy is improved, but computational time increases
Solution Approach 1:
The patent applies partial action by using l1 optimization to achieve sparsity, which provides accurate representation but requires computational time. However, by combining this with regularized least squares and using the sparsity result to select a reduced set of training samples, the system achieves a balance where the computational time spent on sparsity optimization is offset by the reduced complexity of subsequent processing steps.
Data Source
AI summary
A face recognition method is provided to use sparse representation and regularized least squares-based classification on a computing device. The method includes obtaining an image to be recognized as a test sample y and a set of training images of certain subjects as training sample matrix T, obtaining a sparse representation of the test sample and the training samples including an initial estimation of a sparse vector a, and constructing a new face dictionary comprising training samples with non-zero corresponding coefficients in the sparse vector a for the initial estimation. The method also includes obtaining new coefficients by solving a regularized least squares problem based on the constructed new face dictionary, and determining a face identity of the test sample based on minimum class residual calculated by using the new coefficients.