Low-light Instrument Image Enhancement Method Based on Illumination Estimation

Abstract

The captured low-light instrument images suffer from poor visibility caused by low brightness, blurry details, and noise. These images need to be pre-processed to get high-quality images. Low-light image enhancement isabout improving the visibility of the low-light image. Previous image enhancement approaches have improvedthe quality of the low-light image and achieved some results. However, these methods suffer from the problemof heavy computational burden, and the enhanced images have a noise that shows poor visually pleasing results. In this paper, we present a low-light instrument image enhancement approach based on illuminationestimation to solve these problems simultaneously. YOLOv4 is used to detect the instrument image, which can reduce time consumption. The detection section is cropped as the input image for enhancing an image.The illumination of each pixel is estimated by finding the maximum value in RGB channels. Moreover, adesigned optimization function and Gamma correction are employed to optimize the illumination image as thefinal illumination image. Finally, a bilateral filter is introduced to remove noise. Experiments on the low-lightinstrument image are present to demonstrate the effectiveness of the proposed method and its superiority overthe state-of-the-art methods.