Generation of Photorealistic Face Images Using Deep Convolutional Generative Adversarial Networks

In present time, the area of generating photorealistic images using Generative Adversarial Networks (GAN s) has seen considerable interest due to their capability to make high-quality, high-efficient and diverse visual content. This project mainly aims on proceeding the generation of photorealistic face images by addressing existing vulnerabilities in already existing face generation methods. The main objective is to examine and enhance the detection of generated faces by discriminator in GAN sand customizing the generator of a Deep Convolutional GAN model and the evaluation of its efficiency in producing less vulnerable to naked eye, visually clear, and realistic images. By utilizing diverse datasets such as CelebA, this study also aims to develop powerful methodologies that can reliably generate photorealistic facial images. Here the approach involves a thorough examination of the strengths and weaknesses of existing GAN models in generating realistic facial images. Through experimentation with diversified Deep Convolutional GAN architecture and optimization techniques, this study aims to attain significant improvements in image fidelity and realism. By utilizing the large and diverse datasets available, the work intends to traverse various strategies to mitigate common artifacts and inconsistencies observed in GAN -generated faces. Here using different variants of DCGAN s, significant results were obtained indistinguishable from real photographs.