Adaptive Hierarchical Edge Detection: Enhancing Real-Time Artistic Stylization in Computer Graphics

Non-Photorealistic Rendering (NPR) has gained prominence in gaming, offering stylized, artistic expressions over hyperrealism. This paper introduces a unified, real-time post-processing shader architecture that leverages adaptive sensitivity and a three-layer hierarchy to overcome limitations in traditional edge detection algorithms. By employing a Max-pooling fusion technique, our method preserves edge integrity and recovers details in low-light areas, facilitating shorter asset development times. Experimental results demonstrate a significant reduction in manual tuning efforts, achieving high-fidelity artistic stylization with minimal performance overhead across diverse game engines. The complete source code and usage guidelines necessary to reproduce these results are publicly accessible at https://github.com/Chronos-Asteri/AHEAD-AdaptiveEdgeDetection (DOI: https://doi.org/10.5281/zenodo.18765546).