使用奇異值分解方法進行三維彩色物件的材質壓縮與合成之研究

Texture Compression and Synthesis of 3D Color Object Using Singular Value Decomposition

周鑫佑(Hsin-Yo Chou)

指導教授 ： 黃仲陵 ； 張意政

國立清華大學/電機資訊學院/電機工程學系/碩士(2006年)

摘要

數位的虛擬3D合成影像是一門很有趣的課題，相較於傳統的影像合成，更能夠還原真實的呈現方法，並且廣泛地應用在電影特效、遊戲動畫以及虛擬實境上。在未來裡，將虛擬3D影像透過網際網路或平面媒體傳送，更可以達到具豐富地互動式娛樂效果。產生物體虛擬影像的方法中，以影像為基礎和模型為基礎的方法最具有代表性，而這兩者方法都仍有一些缺點，我們結合採用這兩者方法來呈現擬真的3D合成影像。因此，我們不需要知道物體的表面特性，根據取樣的材質和物體的幾何資訊，可以很準確地還原真實物體的風貌。

在本篇論文中，我們在取樣真實物體的外表時，我們取樣的方式是一種view-dependent mapping的方法，物體是處於不同光源環境和具有不同視角的變化。我們提出一個系統架構，利用主成分分析法分解原有取像資料，分解後的係數會再經由三維小波轉換。因此，我們會將取樣下來的材質，會在三維座標下做壓縮，而此其中二維座標平面是定義在3D物體的表面上，另一維是定義在時間軸上。將這些材質在二維平面與時間軸上的多餘不必要的資訊去除。最後係數再透過量化編碼和熵編碼，來達到最小資料量儲存。

這些壓縮後的係數會送入至我們rendering process box裡，可以呈現四種不同結果，分別為1.原有光源方向，還原原有3D Pose，2.在原有光源方向下，合成新的3D Pose，3.在原有的3D Pose下，合成重新打光的效果以及4.合成新的3D Pose，在此新的3D Pose重新打光。另外，根據取樣的材質和物體的幾何資訊，我們可以利用deformation box將物體的表面光場(light field)做切割，將3D物體做變形，變形後部位也能夠正確地呈現光影變化。

關鍵字

影像為基礎呈現 ； 模型為基礎呈現 ； 壓縮 ； 三維小波轉換 ； 主成份分析法

並列摘要

Digital virtual images and synthesis of 3D objects is a very interesting subject. Compared with traditional image composing, this subject really presents original view of 3D objects and is extensively applied to special movie effects, animation, computer games and virtual reality. In the future, delivering virtual 3D images by the way of internet and print media can provide more interactive entertainments plentifully.

In this thesis, we combine the image-based and model-based methods and sample the appearance of an object based on view-dependent mapping method. An object is illuminated under different lighting directions, and variations of appearance are changing under different viewing directions. We propose a system framework to utilize principal component analysis (PCA) to decompose the original captured data. These decomposed coefficients will be compressed by three-dimensional discrete wavelet transform (3D DWT). These appearances are compressed in the 2D coordinate system defined on the 3D model surface. The final remainder data will be further compressed via quantization and entropy coding to achieve minimum storage.

After compressed, these coefficients are delivered to our rendering box and can generated four different applications. They are 1.original pose under original illuminations, 2.novel pose under original illuminations, 3.original pose under novel illuminations and 4.novel pose under novel illuminations.

Based on the model-based method, we can utilize a deformation box to split the surface light field of an object to deform the sub-light field and the deformed parts can correctly present variations due to illuminations.

並列關鍵字

image-based rendering ； model-based rendering ； compression ； 3D wavelet transform ； principla component analysis

參考文獻

[5] Sato Y., Wheeler M. D., and Ikeuchi K. Object Shape and Reflectance Modeling from Observation. . In Proceedings of ACM SIGGRAPH 1997, pp. 379-388.

[9] Nishino K., Sato Y., and Ikeuchi K. Eigen-Texture Method: Appearance Compression based on 3D Model. In Proceedings of Computer Vision and Pattern Recognition 1999, pp.618-624.

[10] MALZBENDER, T.,GELB, D., AND WOLTERS, H. 2001. Polynomial texture

[12] Kristensen A.W., Akenine-Möller T., and Jensen H.W. Precomputed Local Radiance Transfer for Real-Time Lighting Design. In Proceedings of ACM SIGGRAPH 2005.

Representation and Hardware Rendering of Surface Light Fields. In Proceedings of ACM SIGGRAPH 2002, pp.447-456.