Reconstruction Of Structured Microfacet Model Based-on Single Image

With the wide application of physics-based 3D scene rendering technology in the fields of video and audio entertainment,industrial design and other industries,the demand for rendering realism is getting higher and higher.The quality of the rendering result is closely related to the expressiveness of the surface texture material.The key to improving rendering quality is to acquire material resources that can accurately simulate surfaces with complex structures.In physics-based rendering,we use the microfacet model to model the surface of the object as a collection of several microscopic mirrors to simulate the appearance of different materials,and use the normal map to preserve the normal direction of each point on the microfacet.The high quality collection,representation and production of materials have always been the core field of computer graphics,and there are many problems to be optimized in the process of making microfacet models.The traditional microfacet material production requires professionals to manually collect and design,and then use 3D modeling software to generate normal map,which is inefficient and difficult to ensure the quality of map.In the process of practical application and rendering,we found that sometimes the hand-made maps did not conform to the physical laws in real life,and could not meet the realistic rendering requirements of "fake from real".The urgent demand for higher realistic rendering effect promotes the rapid development of material research.Researchers not only focus on material construction and generation,but also focus on reverse high-precision material recovery and reconstruction.recovering material maps from real or synthetic images,predicting the normal distribution model of the material surface,etc.Among them,the general form of material map cannot well represent the highfrequency material surface,and the work of predicting the normal distribution model avoids the explicit modeling of the microfacet.Even if a good visual effect can be obtained,it cannot guarantee the material represented by this normal distribution model is real in the physical world.Therefore,it is necessary to implement a method that can explicitly recover microfacet model of high-frequency material objects.This thesis focuses on the representation and construction of structured microfacet model,and proposes a representation and construction method of structured microfacet model in view of the research blank of parametric representation of microfacet in current field.In this method,the geometric model of structured microfacet primitives is transformed into a parameterized representation,and based on this parameter representation,a series of generation and arrangement formulas of microfacet primitives are designed to explicitly model the microfacet,and finally the normal map representation of the microfacet is obtained.The method proposed in this thesis provides a physics-based parametric representation system for microfacet models and a simple automatic normal map production process for microfacet models,which solves the low efficiency of traditional process.In addition to improving the production efficiency,our method can also guarantee the physical rationality of material resources.Aiming at the problem of reverse restoration of microfacet model,a method of microfacet model reconstruction based on single image is proposed in this thesis.By means of supervised learning,recover microfacet primitive parameters of the object surface material from a single rendered image,and these parameters are used to effectively restore and reconstruct the microfacet model.The reconstructed image can maintain good feature consistency with the input image and can be used as material resource by other rendering engines.