Enhancing Image Classification Performance through Discrete Cosine Transformation on Augmented Facial Images using GANs

Abstract

The continuous advancements in technology are profoundly influencing various domains, including the realm of artificial intelligence. Within this field, the development and training of facial recognition systems have emerged as one of the most prominent research areas. Nowadays, facial recognition systems are rapidly replacing traditional security methods. In order to develop a good face recognition system, the training process must be provided with sufficient data. Recently, the number of open-source data that can help improve the accuracy of face recognition systems is limited. Generative Adversarial Networks (GANs) are a type of machine learning algorithm comprising two interconnected neural networks that engage in a competitive relationship. It is widely used in work domains such as image creation, image manipulation, super-resolution, text visualization, photorealistic images, speech production, and face aging. In the study, the lack of data for training face recognition systems was first solved with synthetic face images obtained with GANs. In the subsequent stage of the investigation, the aim was to enhance the image classification procedure through the application of the discrete cosine transform to the images. This approach aimed to fortify facial recognition systems against the presence of authentic-looking fabricated faces within virtual environments. In the study, it was found that the classification of faces could be improved by 30% compared to the normal classification model. The primary objective of this research endeavor is to make a significant contribution towards the development of highly accurate facial recognition systems.