Adaptive Semantic Priors Guided Multiple Scale Low-Light Image Enhancement

¹ Dept of ECE, Nandha Engineering College Erode, India
² Dept of ECE, Nandha Engineering College Erode, India
³ Dept of ECE, Nandha Engineering College Erode, India
⁴ Assistant Professor, Dept of ECE, Nandha Engineering College Erode, India

This email address is being protected from spambots. You need JavaScript enabled to view it.
This email address is being protected from spambots. You need JavaScript enabled to view it.
This email address is being protected from spambots. You need JavaScript enabled to view it.
This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

To address those major weaknesses of current methods of low-light enhancement, we introduce an Adaptive Semantic Prior Guided Multi-Scale Low-Light Image Enhancement (ASPG-MSLIE) framework which consists of three sections as illustrated in the abstract. In order to resolve this problem we suggest a U-Net backbone with a lightweight semantic prior extractor and a dynamic illumination integration strategy. The network is capable of maintaining the perceptual relevance of high-level semantic context, and suppressing noise in other semantic contexts, through conditioning improvement. So-dimensionality is confirmed by experiments on LOL test set (23.8 dB PSNR, 0.86 SSIM at 0.18 s inferring time) and VE-LOL (88.4% pixel accuracy), confirming state-of-the-art performance over Retinex-Net, KinD, Zero-DCE, and EnlightenGAN.