Font Size: a A A

Research On Modeling And Renderong The Complex Materials

Posted on:2023-05-13Degree:DoctorType:Dissertation
Country:ChinaCandidate:J Q ZhuFull Text:PDF
GTID:1528306617952389Subject:Software engineering
Abstract/Summary:PDF Full Text Request
Realistic rendering is a computer simulation process of the physically-based light propagation in 3D scenes.During rendering,we transform the 3D scenes created by designers,consisting of viewpoints,light sources,3D models,characters,materials,etc.,into highly realistic images through global illumination calculations.Rendering is an important research direction in the field of Computer Graphics.Rendering is widely used in films and animations,product design,advertising design and other industries to provide a variety of gorgeous and realistic visual effects.The continuous emergence of excellent films and animations cannot be separated from the development of realistic rendering technology.More realistic rendering effects will make a qualitative leap in the animation and film works.Photorealistic rendering algorithms are computationally complex,especially for the scenes design of 3D animation and special effects film works which contain a lot of models,materials,lighting,etc.,making rendering the most time-consuming stage.In recent years,with the development of computer hardware and software technology,people expect to see more realistic rendering effects.To achieve this goal,we need to solve some issues such as light transmission,material modeling,and post-processing noise reduction for physically-based rendering.Materials describe the propagation of light reflection,scattering and refraction on an object,and are closely related to the geometric properties and characteristics of the object.Traditional material models tend to describe the geometric and reflective properties of objects in general,which are suitable for expressing simple and regular light reflection distribution.The objects such as glinty surfaces,hair/fur,skin,fabric,complex luminaires,etc.have more complex reflectance which can’t be described by traditional material models.Rich surface details,and these detailed features lead to objects present rich and variable light appearance.We call those materials as complex materials,and study how they scatter light.The problem is how to derive and calculate the reflectance distribution function of such objects.There are two challenging problems need to be solved for modeling and rendering complex materials:because of the small feature size of complex materials,the memory space is large to accurately represent the complex feature,which seriously affects the modeling and representation of complex materials;when moving the camera position,the details of the observed surface are different,we expect to see more details when the camera is close to the object surface,and we expect to achieve reasonable simplification when the camera is pulled away.This thesis conducts an in-depth study of the modeling and rendering algorithms for complex materials to address the above issues.This thesis focuses on modeling and rendering of three common complex materials:glinty surface,hair/fur and complex luminaire.Since glinty surface,hair/fur and complex luminaire can be found everywhere in the real world,accurate rendering of their appearance has a great impact on the rendering results and has important theoretical research and practical application value.The main research work and innovations in this thesis are as follows:1)This thesis proposes a glinty surface modeling and rendering algorithm based on discrete microsurface.This algorithm can accurately and efficiently describe the highfrequency light variations due to discontinuous microsurface distribution,thus completing the microsurface material modeling system.The microsurface distribution of the glinty surface is modeled by using algorithms based on texture synthesis and hierarchical clustering.The normal distribution of the glinty surface is modeled using four-dimensional Gaussian elements,and the approximation to the integral is achieved by accumulating contributing discrete fourdimensional element values during rendering,which accelerates rendering while reducing the storage space.Also,the special characteristics of GPU structure are exploited to optimize GPUbased data representation and computation for real-time glinty surface rendering by designing the accelerate structure which is suitable for GPU rendering.2)This thesis proposes an efficient hair modeling and rendering algorithm based on aggregated fur.In order to accurately model the multiple scattering of the fur,an aggregated fur model the light scattering(including single scattering and multiple scattering)of a bunch of single fur fibers,and the aggregated fur reflection distribution is solved based on neural network prediction.A practical heuristic aggregated fur simplification algorithm is designed to achieve smooth transitions at different observation positions with different detail levels.By using the aggregated fur model,the memory occupied by fur geometry can be reduced and the multiple scattering of fur can be rendered efficiently and accurately.3)This thesis proposes a complex luminaires representation and rendering algorithm based on neural networks.By recording the light transmission inside the complex luminaire on the bounding geometry,the light-field information recorded on the bounding geometry is compressed by the neural network to represent the complex luminaire.By using the bounding geometry and the neural network that records the light-field,we don’t need to keep the real geometry and material of the complex luminaire which successfully reduces the memory space.The appearance of complex luminaires can be directly queried during rendering by using the neural networks,which enables efficient rendering of complex luminaires.In addition,realistic rendering of complex luminaires is achieved by training an importance sampling network to guide how to importance sample complex luminaires.This thesis focuses on the highly realistic rendering of complex materials.The proposed complex material rendering models in this thesis have been integrated into the self-developed rendering system "RWing" of Shandong University and successfully applied to film and animation,virtual reality,digital twins,etc.
Keywords/Search Tags:physically-based rendering, material modeling, neural network, hair/fur rendering
PDF Full Text Request
Related items