三維模型表面紋理的無縫拼接技術(shù)研究
發(fā)布時間：2019-05-27 18:32
【摘要】：空間實體的三維重建,是虛擬現(xiàn)實領(lǐng)域的重要研究內(nèi)容,它在醫(yī)學(xué)、電影、游戲、工業(yè)品造型和可視化以及館藏文物的數(shù)字化保存方面得到越來越廣泛的應(yīng)用。其中,幾何重建和紋理重建是三維重建工作的核心內(nèi)容。幾何重建的目的是獲取模型的幾何結(jié)構(gòu),而紋理重建則賦予三維模型紋理信息。紋理重建主要涉及兩個問題,即紋理圖像與空間幾何模型的配準(zhǔn)和多視紋理之間的無縫拼接。在給定幾何重建模型與多視圖像配準(zhǔn)關(guān)系的條件下,紋理映射主要有兩種表達形式。其中一種是對點云數(shù)據(jù)直接進行處理,點云上某一空間點的顏色取值為它在所有輸入圖像上對應(yīng)點顏色值的加權(quán)平均,其中,權(quán)值取決于輸入圖像所對應(yīng)的觀測方向與該空間點法向的夾角;另外一種做法是,先把點云數(shù)據(jù)轉(zhuǎn)換為多邊形網(wǎng)格結(jié)構(gòu),再對網(wǎng)格的每一個面求取相應(yīng)的紋理貼圖,此時問題的關(guān)鍵在于如何很好地處理紋理接縫。引起紋理接縫不平滑的因素分兩類:一是幾何模型本身不精確、配準(zhǔn)誤差、標(biāo)定誤差以及成像的中心投影畸變等幾何因素;二是因拍攝角度不同、相機感光變化等因素引起的紋理圖像色差,也稱為輻射因素。針對第一個問題引起的圖像錯位,一種做法是引入馬爾可夫隨機場模型,通過圖割、消息傳遞等算法來尋求最優(yōu)解。之后,需要對分割后的各紋理區(qū)塊之間的色差進行處理,比較常用的方法有多波段融合與泊松融合算法。本文利用消費級深度相機和配準(zhǔn)好的多視圖像與網(wǎng)格模型,對多視紋理的無縫拼接方法做了探討,作者完成的主要工作包括:1.提出一種多視影像高光區(qū)域判定策略以及相應(yīng)的去高光方法。首先把各個視角的圖像反投影到可見的空間三角面元上,通過計算并比較同一三角面元處的不同圖像灰度的平均值,根據(jù)單張圖像與其他圖像的偏差值分布情況找出存在高光的圖像。通過提取圖像高光區(qū)域附近紋理的低頻信息,并與原高光區(qū)域進行泊松融合,從而減弱或去除高光。2.提出一種兼顧幾何錯位與顏色偏差的紋理接縫優(yōu)化方法。首先考察了引起紋理接縫不一致現(xiàn)象的可能因素,并采用局部色彩段匹配的方法來求解紋理局部平移量以及顏色補償量,取代標(biāo)簽集擴充方法。利用匹配結(jié)果可以給出紋理在接縫處的錯位信息,同時對接縫紋理顏色進行相應(yīng)補償。從而保證了各個紋理塊之間接縫處紋理的平滑過渡。3.實現(xiàn)了對現(xiàn)有紋理打包算法的優(yōu)化。通過研究若干經(jīng)典的紋理打包算法,并分析、比較它們各自的優(yōu)缺點,給出了一個兼顧存儲空間與渲染效率的實現(xiàn)。相對于傳統(tǒng)的紋理映射方法,本文提出的方法改善了紋理映射的局部細節(jié)表現(xiàn)。
[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.
【學(xué)位授予單位】：中國科學(xué)院大學(xué)(中國科學(xué)院工程管理與信息技術(shù)學(xué)院)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：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.
【學(xué)位授予單位】：中國科學(xué)院大學(xué)(中國科學(xué)院工程管理與信息技術(shù)學(xué)院)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP391.41
【相似文獻】
相關(guān)期刊論文 前10條
1 董e，
本文編號：2486383
本文鏈接：http://sikaile.net/kejilunwen/ruanjiangongchenglunwen/2486383.html
最近更新
教材專著
