다중 렌더링 신호 기반 적응형 디지털 워터마킹: JND 모델과의 통합을 통한 지각적 품질 향상

초록

In digital watermarking for 3D content protection, existing methods fail to exploit intermediate rendering signals and apply uniform embedding strength, leading to inevitable perceptual quality degradation. To overcome this limitation, we propose an adaptive watermarking method that combines rendering signals with a JND model. By analyzing correlations between five rendering signals and JND via Mutual Information, we identify Normal Map (0.073 nats) as the optimal guide signal and design two pixel-wise strength adaptation strategies: Product and Gated. Experiments on 960 samples from six 3D models demonstrate a PSNR improvement of 5.47dB and SSIM enhancement of 0.0059 over the baseline while maintaining 100% message recovery. The computational overhead is 19.4%, making it practical for real-time applications.

키워드