Face Recognition Using The Subspace and Deep Learning Algorithms For Cases of Sufficient and Insufficient Data

In face recognition, the distance criterion significantly influences the recognition rate. Misclassified test signals can be accurately reassigned to the correct class using various distance measures and the nearest neighbor algorithm. This study uniquely explores the recognition performance of DCVA, Fisherface subspace classifiers, and Convolutional Neural Network (CNN) in face recognition, an aspect not thoroughly explored in the literature. Accordingly, this study introduces a Discriminative Common Vector-based (DCVA) algorithm utilizing various distance measures for face recognition for the first time. Additionally, the Fisherface-based algorithm uses different distance measures and nearest neighbors. Experiments were conducted on three different face databases. The images were downsampled to simulate both sufficient and insufficient data conditions. Experimental results indicate that the Correlation distance measure generally outperforms the Euclidean distance for the DCVA and Fisherface-KNN algorithms under both data conditions. The Fisherface-KNN algorithm surpasses the classical Fisherface in performance for various distance measures and nearest neighbor numbers and yields better recognition rates than the DCVA algorithm in sufficient data conditions. Moreover, while DCVA and Fisherface-KNN achieved superior results for two smaller face databases, CNN demonstrated better performance for larger databases.