Low-Light Image Enhancement Using a Simple Network Structure

Abstract

Images taken in light are often hard to see and have poor quality. They can be blurry and noisy. Lacks important details. This is a problem for things like object detection, surveillance, self-driving cars and medical imaging. So making light images better is a very important job in image processing and computer vision.Traditional ways of making images, like histogram equalisation and gamma correction, can make images brighter and have more contrast. They can also make images look too bright, change the colours, and make noise worse. To fix these problems, people have started using learning techniques to make low-light images better.This paper talks about a way to make low-light images better using a simple network. The goal is to make images brighter, have contrast and look better, all while using less computer power. The method uses a convolutional neural network that is designed to learn how to map low-light images to better images. This network is not as complicated as deep learning models so it does not need as much computer power or training data.The network first looks at the light image and pulls out important features like how bright it is, the texture and the colours. Then it uses these features to figure out how to make the image brighter and better. The network is trained using pairs of light and normal light images so it can learn how to make the best enhancements. The network also tries to keep the image looking natural and not too noisy.The results show that this method works well. It makes light images brighter, has more contrast and is easier to see all while keeping noise down. The network is also fast. Does not use too much computer power, which is better than other deep learning models. When we test it using things like Peak Signal to Noise Ratio and Structural Similarity Index Measure, it works well as other methods.Overall, this new way to make light images better is a good solution. It is simple. Does not use too much computer power, so it is good for real-time applications and devices that are not very powerful. This helps make better image enhancement systems for things like object detection, surveillance, self-driving cars and medical imaging. Low-light image enhancement is very important for these applications. This method is a good way to do it. Low-light images can be very hard to work with. This method makes them look better and more useful.