Auto Bayesian Multi-Scale RetineX with Color Restoration of Low-Light image enhancement

This paper presents AutoBMSRCR (Auto Bayesian Multi-Scale Retinex with Color Restoration), a novel method for low-light image enhancement. By combining Retinex theory with Bayesian optimization, AutoBMSRCR adaptively fine-tunes parameters for superior image quality with low computational cost. Using GPU-accelerated bilateral filtering via PyTorch and custom color restoration, the method significantly improves key metrics like UIQM and NIQE, with 9.6% and 64.5% gains, respectively, over competing methods. AutoBMSRCR is ideal for real-time applications on mobile and embedded systems, offering a promising solution for low-light enhancement.