An Improved U-Net Architecture for Low Light Image Enhancement for Visibility Improvement

Images are a significant medium to speak to important data. It might be troublesome for computer vision applications and people to remove important data from images with low light. Presently, the improvement of low-light images is a difficult undertaking in the space of image processing and graphics. In spite of the fact that there are numerous techniques for image improvement, the current strategies frequently produce imperfect outcomes concerning the bits of the image with extraordinary or ordinary light, and such strategies additionally definitely corrupt certain visual relics of the images. Taking an image in low light is challenging due to low contrast, short-exposure images suffer from noise, while long exposure can induce blur and is often impractical. Hence the model should address for low light image enhancement must play out the accompanying assignments: improving contrast, preserving details, noise suppression, and color correction. This paper a deep architecture has proposed to improve the performance of low light images. The developed deep architecture has trained and tested with publicly available dataset namely seeing in the dark dataset. The proposed architecture achieved performance metrics such as Peak Signal to Noise Ratio, Mean Square Error and Structural Similarity Index of 41.22, 0.02 and 0.92 respectively.