Low-Light Image Enhancement Based on Nanoscale Stochastic Magnetic Tunnel Junctions

Images captured in low-light environments suffer from insufficient luminance and lost details, which severely restrict their application in scenarios such as surveillance and aerial photography. Conventional sigmoid-based luminance adjustment relies on software computation and is limited by issues including response latency and poor hardware compatibility. In this work, we propose a low-light image enhancement scheme that leverages magnetic tunnel junctions (MTJs) to implement hardware-based fitting of the sigmoid function. The scheme transforms low-light images from the RGB to the HSV color space and decouples the luminance component from chrominance components, enabling independent luminance adjustment. Utilizing the sigmoid-like relationship between the switching probability of an MTJ and the applied current, the scheme maps low pixel values in the luminance channel to corresponding current amplitudes and directly accomplishes nonlinear fitting of the sigmoid function through the intrinsic physical response of MTJ, which enables efficient enhancement of the luminance component. Experimental results demonstrate that the proposed method effectively improves the luminance and detail clarity of low-light images, achieving an improvement of 3.82 dB in PSNR and 9.91% in SSIM compared with the original sigmoid method and a performance gain of 13.46% in SSIM over the lightweight learning-based scheme. Furthermore, the proposed approach supports the efficient processing of low-light images with different resolutions and is well-suited for resource-constrained edge terminals.This work contributes both theoretical insights and engineering value to the practical deployment of low-light image enhancement technologies.