摘 要

目前，體繪制已成為三維數(shù)據(jù)場(chǎng)可視化的重要技術(shù)手段之一，在科學(xué)計(jì)算和工程領(lǐng)域受到人們的普遍重視和廣泛應(yīng)用。體繪制的傳遞函數(shù)將三維體數(shù)據(jù)的體素值映射成光學(xué)成像參數(shù)，直接決定了三維重建圖像的質(zhì)量。但長(zhǎng)期以來(lái)，體繪制的傳遞函數(shù)的設(shè)計(jì)問(wèn)題一直沒(méi)有得到很好的解決，成為制約體繪制技術(shù)發(fā)展和應(yīng)用的瓶頸，也是近年來(lái)體繪制研究的關(guān)鍵技術(shù)和熱點(diǎn)問(wèn)題。本論文在研究分析當(dāng)前體繪制傳遞函數(shù)設(shè)計(jì)方法的基礎(chǔ)上，利用聚類算法和極端學(xué)習(xí)機(jī)來(lái)指導(dǎo)和優(yōu)化體繪制傳遞函數(shù)的設(shè)計(jì)過(guò)程，以實(shí)現(xiàn)設(shè)計(jì)過(guò)程的智能化和自動(dòng)化。本論文的主要工作如下：
(1)提出了一種基于K均值聚類算法的體繪制多維傳遞函數(shù)設(shè)計(jì)方法，在利用灰度-梯度直方圖分析體數(shù)據(jù)內(nèi)部結(jié)構(gòu)信息的基礎(chǔ)上，應(yīng)用K均值聚類算法對(duì)整個(gè)體數(shù)據(jù)進(jìn)行聚類分類，對(duì)屬于不同聚類中的體素值和不透明度進(jìn)行偽彩色映射，實(shí)現(xiàn)體數(shù)據(jù)與彩色編碼的轉(zhuǎn)換關(guān)系。實(shí)驗(yàn)表明，該方法所設(shè)計(jì)的體繪制傳遞函數(shù)能夠揭示體數(shù)據(jù)的內(nèi)部結(jié)構(gòu)關(guān)系，具有算法簡(jiǎn)潔、計(jì)算效率高、操作方便、重建的三維圖像逼真、質(zhì)量高等優(yōu)點(diǎn)。
(2)在深入分析神經(jīng)網(wǎng)絡(luò)應(yīng)用于體繪制傳遞函數(shù)設(shè)計(jì)的可行性和有效性的基礎(chǔ)上，提出了一種基于極端學(xué)習(xí)機(jī)的體繪制傳遞函數(shù)設(shè)計(jì)方法，將極端學(xué)習(xí)機(jī)應(yīng)用于傳遞函數(shù)的設(shè)計(jì)，并通過(guò)一個(gè)可供用戶交互操作的界面，利用極端學(xué)習(xí)機(jī)對(duì)體數(shù)據(jù)進(jìn)行分類，并對(duì)不同的類賦予不同的顏色值和不透明度，達(dá)到按類進(jìn)行體繪制的效果。實(shí)驗(yàn)表明，該方法設(shè)計(jì)的傳遞函數(shù)能夠有效的分辨不同的物質(zhì)，繪制的圖像清晰，學(xué)習(xí)效率與BP神經(jīng)網(wǎng)絡(luò)相比有大幅提高。

關(guān)鍵詞 體繪制；傳遞函數(shù)；灰度-梯度直方圖；K均值聚類；極端學(xué)習(xí)機(jī)












Abstract

Currently, volume rendering has become one of the important technical methods for three-dimensional data visualization.And it has been widely valued and used in scientific computing and engineering. Transfer function of volume rendering maps voxel value of volume data to the optical imaging parameters, which directly determines the three-dimensional reconstructed images’ quality. But for a long time, the design problem of transfer function of volume rendering had never been satisfactorily resolved, which became the bottleneck of the development and application of volume rendering and has been the key technology and hot issue of volume rendering in recent years. Based on researching and analysising the current method of transfer function design of volume rendering, this thesis used clustering algorithm and extreme learning machine to guide and optimize the the process of transfer function design of volume rendering, which can make the design process intelligent and automatic. The main work of this paper is as follows:
(1) This paper proposed a novel method of multi-dimensional transfer function design of volume rendering. Based on anglicizing the internal structure of volume data by the scalar and the gradient magnitude histogram, all the volume data was classified using K-means clustering algorithm. Then, the volume data belonging to different clustering was pseudo-color mapped for the transformation between volume data and color coding. The experimental results show that transfer function designed by the proposed method can reveal the internal structures of volume data. And our method has the advantages of simple algorithm,high computational efficiency and convenient operation.The reconstructed three-dimensional images are more fidelity and have higher quality.
(2) Based on the depth analysis of the feasibility and effectiveness of using neural network in the field of transfer function of volume rendering, the paper proposed a new method of transfer function design of volume rendering based on extreme learning machine. The method applied extreme learning machine to transfer function design. First all the volume data was classified using extreme learning machine through a interaction user interface. Then, the classified volume data was mapped to different color and opacity.The experimental results show that transfer function designed by the proposed method can effectively separate the different substances,and the reconstructed three-dimensional images are relatively clear. Compared to BP neural network, the learning rate of our method has great increse.
Key words: volume rendering; transfer function; scalar and gradient magnitude histogram; K means clustering; extreme learning machine

目 錄
摘 要 I
Abstract III
第1章 緒論 1
1.1 研究的目的和意義 1
1.2 國(guó)內(nèi)外的研究現(xiàn)狀和發(fā)展趨勢(shì) 3
1.3 論文的主要內(nèi)容和組織結(jié)構(gòu) 4
第2章 體繪制的傳遞函數(shù) 6
2.1 概述 6
2.2 體繪制的傳遞函數(shù) 12
2.2.1 傳遞函數(shù)的定義 12
2.2.2 傳遞函數(shù)的分類 13
2.3 體繪制傳遞函數(shù)的主要設(shè)計(jì)方法 16
2.3.1 手動(dòng)調(diào)節(jié)法 17
2.3.2 圖像中心法 17
2.3.3 數(shù)據(jù)中心法 18
2.3.4 對(duì)象中心法 18
2.4 本章小結(jié) 19
第3章 基于K均值聚類算法的多維傳遞函數(shù)設(shè)計(jì) 20
3.1 概述 20
3.2 灰度-梯度直方圖 20
3.2.1 梯度與物質(zhì)邊界模型 20
3.1.2 灰度-梯度直方圖 21
3.3 K均值聚類算法 23
3.3.1 K均值聚類算法原理 24
3.3.2 K均值聚類算法步驟 25
3.4 基于K均值聚類算法的傳遞函數(shù)設(shè)計(jì) 26
3.5 實(shí)驗(yàn)結(jié)果與分析 28
3.6 本章小結(jié) 30
第4章 基于極端學(xué)習(xí)機(jī)的多維傳遞函數(shù)設(shè)計(jì) 31
4.1 概述 31
4.2人工神經(jīng)..