三维模型表面纹理的无缝拼接技术研究
发布时间：2019-05-27 18:32
【摘要】：空间实体的三维重建,是虚拟现实领域的重要研究内容,它在医学、电影、游戏、工业品造型和可视化以及馆藏文物的数字化保存方面得到越来越广泛的应用。其中,几何重建和纹理重建是三维重建工作的核心内容。几何重建的目的是获取模型的几何结构,而纹理重建则赋予三维模型纹理信息。纹理重建主要涉及两个问题,即纹理图像与空间几何模型的配准和多视纹理之间的无缝拼接。在给定几何重建模型与多视图像配准关系的条件下,纹理映射主要有两种表达形式。其中一种是对点云数据直接进行处理,点云上某一空间点的颜色取值为它在所有输入图像上对应点颜色值的加权平均,其中,权值取决于输入图像所对应的观测方向与该空间点法向的夹角;另外一种做法是,先把点云数据转换为多边形网格结构,再对网格的每一个面求取相应的纹理贴图,此时问题的关键在于如何很好地处理纹理接缝。引起纹理接缝不平滑的因素分两类:一是几何模型本身不精确、配准误差、标定误差以及成像的中心投影畸变等几何因素;二是因拍摄角度不同、相机感光变化等因素引起的纹理图像色差,也称为辐射因素。针对第一个问题引起的图像错位,一种做法是引入马尔可夫随机场模型,通过图割、消息传递等算法来寻求最优解。之后,需要对分割后的各纹理区块之间的色差进行处理,比较常用的方法有多波段融合与泊松融合算法。本文利用消费级深度相机和配准好的多视图像与网格模型,对多视纹理的无缝拼接方法做了探讨,作者完成的主要工作包括:1.提出一种多视影像高光区域判定策略以及相应的去高光方法。首先把各个视角的图像反投影到可见的空间三角面元上,通过计算并比较同一三角面元处的不同图像灰度的平均值,根据单张图像与其他图像的偏差值分布情况找出存在高光的图像。通过提取图像高光区域附近纹理的低频信息,并与原高光区域进行泊松融合,从而减弱或去除高光。2.提出一种兼顾几何错位与颜色偏差的纹理接缝优化方法。首先考察了引起纹理接缝不一致现象的可能因素,并采用局部色彩段匹配的方法来求解纹理局部平移量以及颜色补偿量,取代标签集扩充方法。利用匹配结果可以给出纹理在接缝处的错位信息,同时对接缝纹理颜色进行相应补偿。从而保证了各个纹理块之间接缝处纹理的平滑过渡。3.实现了对现有纹理打包算法的优化。通过研究若干经典的纹理打包算法,并分析、比较它们各自的优缺点,给出了一个兼顾存储空间与渲染效率的实现。相对于传统的纹理映射方法,本文提出的方法改善了纹理映射的局部细节表现。
[Abstract]:3D reconstruction of spatial entities is an important research content in the field of virtual reality. It has been more and more widely used in medicine, film, games, industrial modeling and visualization, as well as the digital preservation of cultural relics. Among them, geometric reconstruction and texture reconstruction are the core contents of 3D reconstruction. The purpose of geometric reconstruction is to obtain the geometric structure of the model, while texture reconstruction gives texture information to the three-dimensional model. Texture reconstruction mainly involves two problems, namely, the registration of texture image and spatial geometric model and the seamless stitching between multi-view texture. Given the relationship between geometric reconstruction model and multi-view image registration, texture mapping is mainly expressed in two forms. One of them is to process the point cloud data directly, and the color value of a space point on the point cloud is the weighted average of the color value of the corresponding point on all the input images, where, The weight depends on the angle between the observation direction corresponding to the input image and the normal direction of the spatial point. In another way, the point cloud data is converted into polygonal grid structure, and then the corresponding texture map is obtained for each surface of the grid. at this time, the key to the problem is how to deal with the texture seam well. The factors that cause the texture seam to be not smooth can be divided into two categories: one is the imprecision of the geometric model itself, the registration error, the calibration error and the central projection distortion of the imaging. The second is the color difference of texture image caused by the different shooting angle and the change of camera sensitivity, which is also called radiation factor. In order to solve the image dislocation caused by the first problem, one method is to introduce Markov random field model to find the optimal solution by graph cutting, message passing and other algorithms. After that, the color difference between the segmented texture blocks needs to be processed, and the common methods are multi-band fusion and Poisson fusion algorithm. In this paper, the seamless stitching method of multi-view texture is discussed by using consumer depth camera and registration multi-view image and grid model. The main work accomplished by the author is as follows: 1. In this paper, a strategy for determining the highlight region of multi-view images and the corresponding highlight removal method are proposed. Firstly, the images of each angle of view are backprojected onto the visible spatial triangulation elements, and the average values of different image grayscale at the same triangulation elements are calculated and compared. According to the distribution of deviation value between single image and other images, the image with highlights is found out. By extracting the low frequency information of the texture near the highlight area of the image and fusing it with the original highlight area, the highlights can be weakened or removed. 2. A texture seam optimization method considering geometric dislocation and color deviation is proposed. Firstly, the possible factors causing the inconsistency of texture seam are investigated, and the local color segment matching method is used to solve the local translation and color compensation of texture, which replaces the label set expansion method. By using the matching results, the dislocation information of the texture at the seam can be given, and the texture color of the seam can be compensated accordingly. Thus, the smooth transition of the texture at the seam between the texture blocks is guaranteed. 3. The optimization of the existing texture packaging algorithm is realized. By studying some classical texture packaging algorithms, analyzing and comparing their advantages and disadvantages, an implementation of combining storage space and rendering efficiency is given. Compared with the traditional texture mapping method, the proposed method improves the local detail representation of texture mapping.
【学位授予单位】：中国科学院大学(中国科学院工程管理与信息技术学院)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
[Abstract]:3D reconstruction of spatial entities is an important research content in the field of virtual reality. It has been more and more widely used in medicine, film, games, industrial modeling and visualization, as well as the digital preservation of cultural relics. Among them, geometric reconstruction and texture reconstruction are the core contents of 3D reconstruction. The purpose of geometric reconstruction is to obtain the geometric structure of the model, while texture reconstruction gives texture information to the three-dimensional model. Texture reconstruction mainly involves two problems, namely, the registration of texture image and spatial geometric model and the seamless stitching between multi-view texture. Given the relationship between geometric reconstruction model and multi-view image registration, texture mapping is mainly expressed in two forms. One of them is to process the point cloud data directly, and the color value of a space point on the point cloud is the weighted average of the color value of the corresponding point on all the input images, where, The weight depends on the angle between the observation direction corresponding to the input image and the normal direction of the spatial point. In another way, the point cloud data is converted into polygonal grid structure, and then the corresponding texture map is obtained for each surface of the grid. at this time, the key to the problem is how to deal with the texture seam well. The factors that cause the texture seam to be not smooth can be divided into two categories: one is the imprecision of the geometric model itself, the registration error, the calibration error and the central projection distortion of the imaging. The second is the color difference of texture image caused by the different shooting angle and the change of camera sensitivity, which is also called radiation factor. In order to solve the image dislocation caused by the first problem, one method is to introduce Markov random field model to find the optimal solution by graph cutting, message passing and other algorithms. After that, the color difference between the segmented texture blocks needs to be processed, and the common methods are multi-band fusion and Poisson fusion algorithm. In this paper, the seamless stitching method of multi-view texture is discussed by using consumer depth camera and registration multi-view image and grid model. The main work accomplished by the author is as follows: 1. In this paper, a strategy for determining the highlight region of multi-view images and the corresponding highlight removal method are proposed. Firstly, the images of each angle of view are backprojected onto the visible spatial triangulation elements, and the average values of different image grayscale at the same triangulation elements are calculated and compared. According to the distribution of deviation value between single image and other images, the image with highlights is found out. By extracting the low frequency information of the texture near the highlight area of the image and fusing it with the original highlight area, the highlights can be weakened or removed. 2. A texture seam optimization method considering geometric dislocation and color deviation is proposed. Firstly, the possible factors causing the inconsistency of texture seam are investigated, and the local color segment matching method is used to solve the local translation and color compensation of texture, which replaces the label set expansion method. By using the matching results, the dislocation information of the texture at the seam can be given, and the texture color of the seam can be compensated accordingly. Thus, the smooth transition of the texture at the seam between the texture blocks is guaranteed. 3. The optimization of the existing texture packaging algorithm is realized. By studying some classical texture packaging algorithms, analyzing and comparing their advantages and disadvantages, an implementation of combining storage space and rendering efficiency is given. Compared with the traditional texture mapping method, the proposed method improves the local detail representation of texture mapping.
【学位授予单位】：中国科学院大学(中国科学院工程管理与信息技术学院)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
【相似文献】
相关期刊论文 前10条
1 董e，
本文编号：2486383
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2486383.html
