Research on UAV nighttime low-light image enhancement based on improved MSR and multiscale guided filtering

Abstract

Aiming at the problems of insufficient brightness, loss of detail information, severe noise interference, and blurred object edges in UAV night low-light scenes, a low-light image enhancement algorithm based on improved Multi-Scale Retinex (MSR) is proposed. First, the input RGB image is converted into HSV color space, and only the luminance V channel is processed to avoid color distortion. Second, multi-scale gradient guided filtering is used to replace the Gaussian filtering in traditional MSR to estimate and fuse illumination components at different scales, thereby improving structure preservation ability. Then, based on Retinex theory, the reflectance component is separated and noise is suppressed by combining gradient-domain denoising methods. Finally, adaptive Gamma correction is used to enhance overall brightness and contrast. Experimental results show that compared with traditional MSR, RetinexNet, and dark channel prior methods, the proposed method achieves excellent performance in peak signal-to-noise ratio (PSNR), structural similarity (SSIM), information entropy (Entropy), and average gradient (AverageGradient), and can significantly improve UAV night target visibility and image visual quality.