Improving CCTV footage in Computer Vision Using Deep Learning Techniques

Abstract

The fields of computer vision and image processing, which deal with the analysis and modification of visual data like pictures and videos, are closely related. While computer vision focuses on tasks like object recognition and localization, image processing is dedicated to enhancing image quality through techniques like filtering and sharpening. Both are crucial as image quality impacts accurate analysis in various applications. Super-Resolution Generative Adversarial Networks (SRGANs) are a significant advancement in image processing, addressing the enhancement of low-resolution images into high-resolution ones. By employing an SRGAN framework with a generator, discriminator, and VGG19 feature extractor, remarkable super-resolution is achieved. SRGANs are valuable for improving image resolution in areas like scientific imaging, surveillance, and digital media. Higher resolution images enable finer detail detection, better object recognition, and enhanced visual experiences. The designed SRGAN demonstrated 85% training and 83.45% validation accuracy, with a testing accuracy of 80% on 1000 test images. Additionally, it achieved 84.55% testing accuracy on a low-resolution image dataset. This framework holds promise for various applications requiring superior image quality and resolution.