Efficient Acquisition And Reconstruction Methods For Bidirectional Texture Function (BTF) Images

Photorealistic graphics has been widely used in the fields such as virtual reality,film special effects production and computer aided design.Photorealistic graphics is dedicated to synthesize photo-realistic images on the computer.In order to achieve this,it is necessary to acquire the texture images of real materials under different viewing directions and lighting directions.These texture images can be called as the Bidirectional Texture Function(BTF)data.Due to the high dimension of the BTF data,there exists two issues in the data acquisition:1)the acquisition time is very long,and 2)the acquired data size is very large.These two issues make it difficult to use the BTF data in the practical applications.The existing literatures have made great contributions to the BTF data in compression,synthesis and rendering.However,the works coping with the efficient acquisition problem are quite limited.This thesis investigates the problems of efficient acquisition and reconstruction of the BTF data(images).The main contributions are summarized as follows:1.A method is proposed to approximately reconstruct the BTF data of materials from angularly sparse measurements.In this method,we automatically cluster a training set and obtain representation bases for these clusters.Based on these representation bases,the sparse sampling positions are obtained using optimal experiment design.Considering that the proposed method can select the acquisition positions of viewing and lighting directions respectively,it can efficiently reduce the numbers of both camera and light sources.2.A fast BTF image super-resolution(SR)algorithm is proposed to improve the resolution of BTF data.The algorithm uses singular value decomposition(SVD)to separate the collected low-resolution(LR)BTF data into intrinsic textures and eigen apparent bidirectional reflectance distribution functions(eigen-ABRDFs),and then improves the resolution of the intrinsic textures via image SR.The HR BTFs can be finally obtained by fusing the reconstructed HR intrinsic textures with the LR eigen-ABRDFs.3.A novel framework for efficient spectral BTF acquisition and reconstruction is proposed.The framework acquires the entire RGB BTF images and just one spectral image of the material.By exploring the characteristics of the BTF imaging system and the spectral imaging system,we find the relationship between the mean BTF image and the spectral image.The full spectral BTFs are reconstructed by fusing the RGB and spectral images based on nonnegative matrix factorization(NMF).