Wavelet Subspace Based Integrated Face Recognition Scheme

Wenhui Li, Ning Ma, Zhiyan WangInternational Journal of Signal Processing Image Processing and Pattern Recognition 2014 · DOI

TL;DR

This paper introduces an integrated face recognition system that combines wavelet decomposition with 2DPCA and fuzzy 2DLDA. By leveraging the strengths of each component, the proposed method aims to improve recognition accuracy and efficiency, outperforming existing individual methods on standard face datasets.

What Problem Does This Paper Address?

Face recognition is a challenging task in computer vision, with existing methods often facing limitations. Traditional approaches like PCA and LDA require converting images into vectors, which is computationally expensive and can lead to high dimensionality. While 2D extensions like 2DPCA and 2DLDA address this by directly processing image matrices, they still have individual drawbacks. 2DPCA excels at extracting general features but is slower for classification, whereas 2DLDA is faster for classification but requires more training time. Furthermore, many wavelet-based methods discard high-frequency information, negatively impacting recognition rates. This paper seeks to overcome these limitations by creating an integrated system that combines the advantages of these techniques and utilizes all frequency sub-bands from wavelet decomposition.

Methods

The proposed "Wavelet Subspace Based Integrated Face Recognition Scheme" is built upon several key components:

  • Wavelet Decomposition: The initial step involves applying a multi-level wavelet decomposition to face images. This breaks down the images into different frequency sub-bands (low and high frequency components in horizontal and vertical directions). The paper specifically mentions a three-layer decomposition.
  • 2DPCA (Two-Dimensional Principal Component Analysis): This method is used for feature extraction. Unlike traditional PCA, 2DPCA directly constructs covariance matrices from image data, reducing computational complexity. It extracts general feature information from the face images without prior knowledge of class labels.
  • Fuzzy 2DLDA (Two-dimensional Linear Discriminant Analysis): This technique is employed for feature selection. It builds upon 2DLDA by incorporating fuzzy set theory. This allows for better consideration of sample structure and distribution information, especially for overlapped samples, leading to more efficient classification. Fuzzy 2DLDA aims to improve classification accuracy and robustness.
  • Nearest Neighbor Classification: A standard nearest neighbor algorithm is used for the final classification step, comparing feature vectors of test images against those of training samples.
  • Integration and Vote Method: The paper describes an integration strategy where features from different frequency sub-bands (obtained after wavelet decomposition) are processed by 2DPCA and fuzzy 2DLDA. The final classification is achieved by combining the results from these classifiers using a vote method.

Datasets

The paper evaluates the proposed method on three commonly used face image databases:

  • ORL Face Database: Contains 400 images from 40 individuals, with 10 images per person. Image resolution is 92x112.
  • YALE Face Database: Contains 165 images from 15 individuals, featuring variations in illumination and expressions. Image resolution is 320x243.
  • FERET Face Database: Contains a large number of images (1400 from 200 individuals, 7 images per person). For this experiment, a subset of 30 individuals was selected. Image resolution is 80x80.

Key Insights and Ideas

  • Complementarity of 2DPCA and 2DLDA: The paper highlights the synergistic relationship between 2DPCA and 2DLDA. 2DPCA is good for general feature extraction without class labels, while 2DLDA excels at classification and utilizes class information. Combining them addresses the weaknesses of each.
  • Leveraging All Frequency Bands: A significant insight is the decision to utilize all frequency sub-bands from wavelet decomposition, rather than just the low-frequency components. This approach is claimed to improve recognition accuracy by not discarding potentially useful high-frequency identification information.
  • Fuzzy Set Theory for Robustness: The integration of fuzzy set theory into 2DLDA is presented as a way to enhance robustness and handle uncertainties or variations in face images more effectively.
  • Integrated Approach for Efficiency: The core idea is that by combining these techniques and processing different frequency components, the overall system can achieve better recognition rates while also reducing computational time compared to using individual methods or less comprehensive approaches.

Key Results

The experimental results, presented in Tables 1, 2, and 3, demonstrate the superiority of the proposed integrated method over several baseline methods:

  • ORL Database: The proposed method achieved an average identification rate of 96.64% with 6 test samples, outperforming Method 1 (2DPCA, 90.01%), Method 2 (2DLDA, 93.52%), and Method 3 (2DPCA+2DLDA, 94.54%).
  • YALE Database: The integrated method reached an average identification rate of 97.89% with 8 test samples, surpassing Method 1 (92.06%), Method 2 (95.36%), and Method 3 (97.06%).
  • FERET Database: On this dataset, the proposed method achieved 95.53% with 6 test samples, exceeding Method 1 (92.01%), Method 2 (86.54%), and Method 3 (95.03%).

The paper also states that the integrated method improves recognition efficiency and is robust to illumination changes.

Limitations and Future Work

  • Limitations: While the paper claims improved efficiency, the specific computational gains in terms of training and classification time are not explicitly quantified with concrete numbers in the provided text. The paper also mentions that some previous wavelet-based methods discard high-frequency information, implying that this is a common limitation that their method addresses.
  • Future Work: Not explicitly discussed in the provided text. However, the success of the integrated approach suggests potential for further exploration of different wavelet bases, more advanced fuzzy logic integration, or combining with other feature extraction techniques.

Summary updated

Abstract

In this paper, based on the study of the Two-Dimensional Principal Component Analysis (2DPCA), Two-Dimensional Principal Component Analysis (2DPCA) and fuzzy set theory, we propose a integrated face recognition algorithm based on wavelet subspace. This method can make good use of the advantages of each single method, and also can make up for the defect of each other. The comparison of the results of the different methods identification effect on the ORL、YALE and FERET face database show, the integrated method proposed in this paper improves the recognition rate, and it also reduces the training and classification time as well.
BibTeX
@article{li2014wavelet,
  author = {Wenhui Li and Ning Ma and Zhiyan Wang},
  title = {{Wavelet Subspace Based Integrated Face Recognition Scheme}},
  journal = {International Journal of Signal Processing Image Processing and Pattern Recognition},
  year = {2014},
  doi = {10.14257/ijsip.2014.7.6.32},
}
Figure