Improving Deep Learning for Face Verification Using Color Histogram Equalization Data Augmentation

Abstract

This paper proposes a new method of improving face verification learning using color histogram equalization by incorporating the results of deep convolutional neural networks (CNNs). The entire process of face verification using deep learning and color histogram equalization is described in detail. This research uses advanced deep learning methods for face verification tasks. Because the CNN achieves the best results for larger datasets, the main challenge is to increase the smaller dataset enhancements and validate in environments different from the training datasets. This paper presents a new training enhancement method. When the face-image datasets were small, we could use our enhanced method to expand the dataset and improve the accuracy of face verification to adopt different environments. Consequently, the accuracy reached 99.7996%, i.e., approximately 7% higher than the result trained on the original dataset.