Wavelet Transformation and Local Binary Pattern for Data Augmentation in Deep Learning-based Face Recognition

Facial recognition is used in many fields such as verification, communication, and other fields. The success of facial analysis is influenced by noise from environmental factors such as illumination, expression, posture, occlusion, and others. Therefore, a solid and effective analytical model is needed to overcome noise using feature extraction and deep learning. This study compares the effect of various wavelet transform methods and local binary pattern (LBP) on facial recognition as additional data on the CNN architecture and the pre-trained VGG16 model on the Yale-B facial recognition dataset. The results showed that the LBP and IDWT features of discrete wavelet transform (DWT) as augmented data resulted in the highest accuracy value of 99.69%.