Generating New Human Faces and Improving the Quality of Images Using Generative Adversarial Networks(GAN)

Abstract

In recent years, deep learning models have gained popularity for producing realistic Images. Recent advancements in computer vision, particularly in deep generative models like GANs, have shown promise in synthesizing realistic images automatically. GANs use a competitive process involving two networks: a generative network and a discriminative network. The discriminative network determines whether an image is real or fake whereas the generative network generates artificial images. The generative network gains the ability to create more convincing images as training goes on in order to deceive the discriminative network. This research study intends to develop novel, high-resolution images of human faces by combining DCGAN (Deep Convolutional Generative Adversarial Network) with ESRGAN (Enhanced Super-Resolution Generative Adversarial Networks). DCGAN is a type of GAN that uses convolutional neural networks in both the generator and discriminator. The generator network learns to produce images from random noise, while the discriminator network learns to differentiate between real and fake images. Further, this study has used the CelebFaces Attributes Dataset (CelebA) to train the proposed DCGAN model, and the Structural Similarity Index (SSIM) to quantitatively evaluate the quality of the generated images. Additionally, ESRGAN is employed to improve the quality of the generated images. The obtained results reveal that combining DCGAN with ESRGAN produces high-quality human faces with clear details and improved resolution.