This chapter focuses on applying deep learning to improve low-light photos,
specifically a Generative Adversarial Network (GAN). Real-life photographs often
exhibit noise, low contrast, and dull hues, particularly when the lighting conditions are
poor. Due to these issues, the images are challenging to comprehend and utilize for
various purposes, including surveillance, medical imaging, and photography. This
study's approach combines deep learning with image processing methods to enhance
image quality. The GAN model is trained to enhance brightness and contrast, resulting
in crisper and more detailed images. These methods first preprocess the images to
remove dark areas and normalize contrast. Three important metrics, PSNR (peak
signal-to-noise ratio), MSE (mean squared error), and SSIM (structural similarity
index), are used in the study to evaluate image improvement using the VE-LOL
dataset. These metrics enable a more accurate assessment of the improvement in image
quality between the enhanced images and the initial low-light images. Since the GAN
learns to recognize patterns in both bright and dark images, it can produce high-quality
images from low-light sources. The results show that the deep learning-based method
outperforms traditional picture-enhancing techniques in terms of clarity and detail.
High-quality images are crucial in various fields, including photography, medical
imaging, and surveillance, where this technology is often applied, by automatically
enhancing the visual quality of photos taken in low-light conditions. This strategy can
assist professionals in a range of fields in taking better and more informative images,
especially in low-light settings.
Keywords: Deep learning, Feature extraction, Feature Refinement, Generative adversarial network, Image enhancement, Low light images.
