Enhanced Image Vision and Resolution during Low Light Conditions using GANs

Abstract

Low light computer vision is an arduous task because of the low signal to noise ratio and less photon count. This means that the images captured in low light experience noise, which can result in blurring of the image. Although there are multiple techniques to overcome the noise and blur, their results are bounded in undue conditions as in there is a drop in the video imaging at night. This low light enhancement is a daring task as there are multiple factors like brightness, de-noising, de-blurring, contrast must be handled at the same time. Even the development of a CNN has proved to perform poorly on such data. This paper uses a technique to take care of this issue using GANs. Our technique gives a platform to enhance the image captured in low light and increase its resolution giving out an enhanced super resolute image. To support the low light image processing, we have used a dataset of low-light images. This method can give promising results on the dataset, and display a break for the future work.