Abstract
Image capturing has been improved over the years, leading to increased usage in various applications for achieving a digital image. The images help keep a memory of the occasion and are also used to achieve an effective and useful means of capturing and studying the different phenomenon in a variety of fields. One of the challenging issues in image capturing is the utilization of limited and underpowered hardware in remote locations for capturing low-resolution images. The applications in such an environment are limited by the hardware, which results in low quality and low-resolution images that cannot be effectively used without improvements through image processing.Due to increased research on this topic, there has been a considerable improvement in the imaging approach. Improving image quality can be achieved through the use of neural networks and deep learning techniques. In this project, we first enhanced the light levels of the images by using Low Light Convolutional Neural Network (LLCNN). After that, Generative Adversarial Networks (GANs) are used for enhancing the image resolution and quality of the images.
The proposed method is a hybrid approach implemented through the effective use of the Low Light Convolutional Neural Networks (LLCNN), Super-Resolution Generative Adversarial Networks (SRGAN), and a custom CNN architecture to produce images with better lighting and resolution. After this processing, the images will be scaled 4x from corresponding low-resolution images, and an image with improved visual perception and fidelity is produced. The results are validated through extensive experimentation, which shows the method's effectiveness to improve images based on PSNR and SSIM scores.