Study Of Real-Time Microstructure Representations And Rendering Methods

The phenomenon of highlight glint can be seen everywhere in the real world,and an accurate reproduction of this visual effect can significantly enhance the realism of the rendering.The microstructure of the material itself generates the glint effect.When light interacts with the microstructure,the energy presents a highly complex irregular distribution in the spatial domain.Some microstructures show glints and other visual effects at specific azimuths under sharp light sources,and different types of microstructures exhibit completely different high-frequency detail characteristics.The use of smooth statistical distribution in conventional computer graphics to simulate the overall characteristics of microstructures leads to the loss of highfrequency detail features in the results.It shows only excessively perfect and smooth highlights.In recent years,a microstructure rendering method based on normal mapping has been widely used in the offline rendering field,which can accurately produce appearance effects such as glints and scratches.However,defining microstructure normal using ultra-highresolution normal mapping brings a significant memory overhead and computational burden,which is difficult to be directly ported to real-time application scenes.The current real-time rendering methods only model a single type of microstructure.Therefore,the pervasive representation and rendering of microstructures in real-time scenes is an important issue that needs to be addressed.This paper proposes two algorithms to fill the gap in microstructure rendering methods in the real-time rendering field.This paper first proposes a real-time rendering algorithm based on MIP-map,which can generate microstructure materials at arbitrary scales while greatly reducing memory overhead and is applicable to most microstructure types.The algorithm first inputs small-scale normal map samples,which generate large-scale high-resolution maps to represent microstructure implicitly.It organizes four-dimensional position-normal Gaussian lobes using the MIP-map hierarchy.We also propose an algorithm that makes full use of the high parallelism of the GPU to efficiently query the Gaussian lobes to fit the approximate normal distribution function quickly and achieves global illumination effects by using light blending and spatial filtering in the real-time ray tracing pipeline and significantly reduces computation and storage burden while maintaining stable high-frequency features in the spatial-temporal domain,and achieves the first multi-scale consistent real-time rendering of microstructure.This paper then proposes a Primitive-based microstructure representation and rendering method,which innovatively assumes that the macroscopic surface of a material is a unified whole composed of a large number of primitive microstructures.Many self-similar materials in the real world can be described by independent geometric primitives,thus enabling the representation of various microstructures.The algorithm increases the diversity of primitives through the spatial transformation of primitives and spatial distribution.At the same time,only a tiny amount of information needs to be maintained,thus significantly reducing the memory overhead,and proposes an analytical formula for computing normal distribution function,which can restore the detailed features of materials with less memory and computational overhead.Unlike dealing with the global microstructure,the algorithm only requires precomputation for the primitive,resulting in significantly shorter preprocessing time and more excellent reusability of the data.Both algorithms have the features of solid microstructure representation,a wide range of representations,and friendly memory and computational overheads.They can be directly applied to real-time rendering engines to reconstruct high-frequency features of materials,such as glints,scratches.The material ranges of the two algorithms are slightly different,as the Primitive-based algorithm is suitable for microstructures with discrete primitive features.In contrast,the MIP-map-based algorithm suits material types with high geometric continuity requirements.