AHEAD: Adaptive Hierarchical Edge Detection for Real-Time Artistic Stylization

There has been a notable increase in the usage of Non-Photorealistic Rendering (NPR) ingames, like the Borderland series and The Walking Dead, which depicts a stylized, artistic expression instead of hyperrealism. Consequently, modern game development has seen an increaseddemand for the creation of ‘ink style’ or ‘comic-book look’ assets characterized specifically by theirbold and clear silhouettes, and shading effect on the surface texture. This has primarily beenachieved by basic edge detection algorithms which result in a noisy image, discontinuous edgesand loss of structural detail. Therefore, the assets are individually edited to get the desired results,leading to increased development time. This paper proposes a unified, real-time post-processingshader architecture that makes use of adaptive sensitivity and a three-layer hierarchy in orderto overcome these constraints. The suggested framework uses a Max-pooling fusion technique tointegrate the scene after breaking it down into separate Silhouette, Structure and Texture layers.This method preserves edge integrity better than conventional linear averaging of the single-passfilters, while also recovering details in low-light areas. The resulting visualization offers the possibility of shorter asset development times by accurately reproducing a hand-drawn artistic style.The full source code and shader files are available online.