Face Recognition Using Principal Component Analysis with Euclidean Distance and Artificial Neural Network Classification Methods

Mahardhika Pratama, Khoirin Nisa, La Zakaria, Mona Arif MudaJournal of Modern Applied Statistical Methods 2024 · DOI

TL;DR

This paper explores face recognition by combining Principal Component Analysis (PCA) for dimensionality reduction with two classification methods: Euclidean distance and Artificial Neural Networks (ANNs). The study demonstrates that a multi-layer ANN with a (4,2) configuration achieves the highest accuracy (91%) and lowest error rate (7.5303%) on the AT&T Laboratories Cambridge face dataset, outperforming both Euclidean distance and simpler ANN architectures.

What Problem Does This Paper Address?

Face recognition, a crucial biometric technology, faces challenges due to the high dimensionality of image data and the inherent variability in human faces (lighting, pose, expression). Traditional methods often struggle with computational complexity and accuracy. This paper addresses the problem of efficiently and accurately recognizing faces by leveraging PCA to reduce the dimensionality of image data, making it more manageable for subsequent classification. The goal is to find the most effective classification method among established techniques for this reduced-feature representation.

Methods

The core methodology involves a two-stage process:

  1. Dimensionality Reduction with Principal Component Analysis (PCA):
    • The raw image data is transformed into a matrix.
  2. PCA is applied to this matrix to extract the most significant features, known as "eigenfaces." This process reduces the number of dimensions (pixels) while retaining most of the variance in the data.
  3. The paper details the calculation of average faces, eigenfaces, and variance proportions, illustrating how PCA identifies the principal components that capture the most information.
  4. Classification:
    • Euclidean Distance: This standard method calculates the distance between the projected test data and the projected training data (eigenfaces) in the reduced-dimensional space.
  5. Artificial Neural Networks (ANNs): Both single-layer and multi-layer ANNs are employed. The paper investigates different configurations of hidden neurons (e.g., single-layer with 2, 3, or 4 neurons; multi-layer with (3,2), (3,3), or (4,2) neurons). The ANNs learn to classify faces based on the PCA-derived features.

The study uses R software for data processing and implements a standard train-test split for evaluation.

Datasets

The research utilizes the AT&T Laboratories Cambridge face dataset. This dataset comprises 400 grayscale images of 40 individuals, with each person having 10 different expressions. The images have a resolution of 27x25 pixels and are represented with 256 levels of black and white color, scaled between 0 and 1. The data was split into training and testing sets, each containing 50% of the data points.

Key Insights and Ideas

  • PCA as an Effective Preprocessing Step: The paper reinforces the utility of PCA in face recognition by demonstrating its ability to reduce high-dimensional image data into a more manageable set of features (eigenfaces) without significant loss of discriminative information.
  • ANNs Outperform Euclidean Distance: The results consistently show that ANNs, particularly multi-layer architectures, achieve higher accuracy and lower error rates compared to the Euclidean distance method for face classification on the PCA-transformed data.
  • The Importance of ANN Architecture: The study highlights that the specific architecture of an ANN (number of layers and neurons) significantly impacts its performance. The multi-layer ANN with a (4,2) configuration emerged as the most effective.
  • Trade-off Between Iterations and Error: Table 1 reveals that while some ANN models (like Single-4) require a very large number of iterations, this doesn't necessarily translate to the lowest error. The multi-layer ANN (4,2) achieved a low error with a substantial, but not excessively high, number of iterations.

Key Results

  • Euclidean Distance Accuracy: Achieved an accuracy of 89%.
  • Single-Layer ANN Accuracy: Ranged from 89% to 90.5%.
  • Multi-Layer ANN Accuracy:
    • ANN(3,2) achieved 89.5% accuracy.
  • ANN(3,3) and ANN(4,2) both achieved the highest accuracy of 91%.
  • Lowest Error Rate: The multi-layer ANN (4,2) achieved the smallest error value of 7.5303.
  • Conclusion: The multi-layer ANN (4,2) is identified as the best performing model due to its superior accuracy and lowest error rate.

Limitations and Future Work

Limitations: * The study uses a relatively small and older dataset (AT&T Cambridge). The performance of the models might differ on larger, more diverse, or more challenging datasets with variations in pose, illumination, and expression. * The paper focuses on a specific set of ANN architectures; exploring a wider range of ANN designs or hyperparameter tuning could potentially yield even better results.

Future Work: The authors suggest extending this study by exploring other image classification methods, such as:

  • Fisherfaces (a variation of PCA)
  • Linear Discriminant Analysis (LDA)
  • Local Binary Patterns (LBP)
  • Histograms of Oriented Gradients (HOG)

This would provide a more comprehensive comparison of different feature extraction and classification techniques for face recognition.

Summary updated

Abstract

Principal component analysis is a multivariate statistical method used for reducing data dimension which can be applied in various fields. One of them is in computer science application, i.e. dimension reduction of image data for face recognition. This research focuses on obtaining average faces matrix, eigenfaces, and data projection results based on principal component scores. The results were then used for further step in face recognition namely classification. Here two classification methods are used, namely Euclidean distance and artificial neural networks (ANN) with single and multi-layers. The data used are from AT&T Laboratories Cambridge collections in April 1992 - April 1994, each line of the data contains pixel from a single image that has 256 levels of black and white color between 0 and 1. Based on the analysis results, the accuracy rate of face recognition using the Euclidean distance method is 89%. At the same time, single-layer ANN produce the accuracy of 90.5% for two hidden layers, 89% for 3 and 4 hidden neurons, while multi-layer ANN produce the accuracy of 89.5% for hidden neurons (3.2), and 91% for hidden neurons (3,3) and (4,2). However, the smallest error was obtained by multi-layer ANN with hidden neurons (4,2) which resulted the error value of 7.5303. Thus we conclude that the multilayer ANN (4.2) outperformed the others and then is chosen as the best classification for the face recognition analysis of the data.
BibTeX
@article{pratama2024face,
  author = {Mahardhika Pratama and Khoirin Nisa and La Zakaria and Mona Arif Muda},
  title = {{Face Recognition Using Principal Component Analysis with Euclidean Distance and Artificial Neural Network Classification Methods}},
  journal = {Journal of Modern Applied Statistical Methods},
  year = {2024},
  doi = {10.56801/jmasm.v23.i2.5},
}
Figure