A Swin-Transformer-Based Network for Adaptive Backlight Optimization

To address luminance discontinuity, halo artifacts, and insufficient temporal stability commonly observed in Mini-LED local dimming systems, this paper proposes an adaptive backlight optimization network, termed SwinLightNet, based on a hierarchical attention mechanism. From a luminance modeling perspective, the proposed method exploits multi-scale feature correlations to achieve spatially smooth and content-adaptive backlight distributions, while incorporating temporal luminance constraints to enhance stability in video scenes. Experimental results demonstrate that SwinLightNet consistently outperforms conventional local dimming algorithms and representative learning-based methods in terms of PSNR, SSIM, and subjective visual quality, validating its effectiveness for Mini-LED backlight optimization.