双向纹理函数（BTF）图像的高效采集与重建方法

发布时间：2018-02-26 16:25

本文关键词： 双向反射分布函数双向纹理函数稀疏采集图像超分辨率光谱重建　出处：《浙江大学》2017年博士论文　论文类型：学位论文

【摘要】：在虚拟现实、电影特效制作和计算机辅助设计等领域中,真实感图形学得到了广泛的应用。真实感图形学致力于在计算机上合成出具有照片真实感的图形,为了达到这个目的,需要采集真实物体表面在不同视角方向和光照方向下的纹理图像。这些纹理图像的集合可以被称为双向纹理函数(Bidirectional Texture Function,BTF)数据。由于BTF数据维度较高,现实采集中存在采集时间过长、采集数据量庞大等问题,给其实际应用带来了困难。现有文献在BTF数据压缩、合成和渲染方面取得了很好的进展,但在高效采集和重建方面的研究工作较少。本论文针对BTF数据(图像)高效采集和重建这两个方面问题开展了以下创新性的研究工作:1.提出了一种从物体样品稀疏采集的BTF数据中重建完整BTF数据的方法。在训练阶段,将训练数据进行聚类,通过数据分解得到能够表征每个类的基函数矩阵,并采用优化实验设计法选出稀疏采集的角度;在采样/重建阶段,只在被选择的稀疏采样角度下采集新物体样品的数据,通过最小二乘法重建出完整的BTF数据。该方法能够分别选择相机和光源的稀疏采集位置,有效地减少了实际所需的相机和光源个数。2.提出了一种快速的BTF图像超分辨率重建的方法。通过奇异值分解将已采集到的低分辨率BTF数据分解为本征纹理矩阵和特征ABRDF矩阵,然后采用图像超分辨率重建算法,提升本征纹理矩阵的分辨率。最后,高分辨率的BTF图像可以通过高分辨率本征纹理矩阵和低分辨率的特征ABRDF矩阵相乘获得。3.提出了一种有效的光谱BTF数据采集和重建的方法。该方法首先获取物体样品完整的彩色(RGB)BTF图像和一幅正视角相机下的光谱图像。然后分析BTF采集装置结构和光谱成像装置结构的特点,建立起两者之间的关联,并得到与光谱图像相关的RGB图像。最后采用基于非负矩阵分解的光谱重建方法,将RGB和光谱图像融合,得到物体样品的光谱BTF数据。
[Abstract]:In the field of virtual reality, film special effects and computer aided design, realistic graphics has been widely used. Realistic graphics on a computer dedicated to synthesize photorealistic graphics, in order to achieve this goal, need to collect real objects on the surface of different view direction and light texture image direction. Set the texture image can be called bidirectional texture function (Bidirectional Texture, Function, BTF). The BTF data of high dimension data acquisition time is too long, the reality of the existence of the collection, huge amount of data collection, for its practical application difficult. The existing literature on BTF data compression, synthesis and rendering progress very good, but in terms of efficient acquisition and reconstruction work less. In this paper BTF (image), data acquisition and reconstruction of the two aspects of the problem The research work has been carried out the following: 1. we proposed a reconstruction method of complete BTF data from BTF data object samples of sparse data. During the training phase, the training data can be obtained by clustering the basis function matrix characterization of each class by data decomposition, and the optimal experimental design method to choose the angle of sparse data in the sampling / reconstruction phase; only the sampling angle collected under the new object data in sparse samples is selected, a complete BTF data by the least squares method. The method can select the reconstruction of sparse data position of the camera and the light source, effectively reduce the actual requirement of the camera and the light source number.2. proposed a super resolution reconstruction of BTF images. The rapid decomposition of low resolution BTF data have been collected into the intrinsic matrix and texture characteristics of the ABRDF matrix by the singular value, then the image Super resolution reconstruction algorithm, enhance the intrinsic matrix texture resolution. Finally, the characteristics of ABRDF matrix multiplication BTF high resolution image by high resolution eigen matrix texture and low resolution.3. put forward an effective method of spectral BTF data acquisition and reconstruction. The method first obtains the object sample full color (RGB BTF) spectral image image and a positive view of the camera. Then analyzing the characteristics of the BTF acquisition device structure and spectral imaging device structure, establish the relationship between the two, and are related with spectral image RGB image. Finally using spectral reconstruction method based on non negative matrix factorization, RGB and spectral image fusion BTF, get the spectrum data of the object samples.

【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】