光線變化及其對視覺問題之應用

光線變化對於物體的外貌扮演了重要的角色，它造成了外觀的濃淡、反光、及陰影的效果。雖然這個世界因能有如此多樣的光影效果而迷人，這些光影效果卻對許多電腦視覺的問題造成了重大的困難。若試著開始描述光線變化所造成的影響，通常會因它的無窮自由度而感到棘手，儘管此問題的本質看似令人怯步，仍有如球面調和函數這樣顯著的進展被提出。球面調和函數提供了一種一致的數學架構來描述複雜的光線變化。反射出的光線可被看成是入射光線和物體表現反射屬性的一種迴旋積的結果，事實上，一個漫射的物體表面可以被當作是作用在入射光訊號上的一個低頻濾波器，畢竟一個漫射的物體即使在複雜的光線上仍能保有平滑的外觀。若將這些概念正式地用訊號處理的觀點來描述的話，就可導致許多新潁有趣的洞悉及在電腦視覺問題上的實際應用，例如辨視、光線估測、以及形狀重建。在本篇論文中，基於以上的技術我們檢視了幾個典型的電腦視覺問題並提出了一些新奇並有效的方法，先前的方法通常都有些踞限的假設，它們通常簡化甚至忽略光線變化所造成的影響，而我們提出的方法即使在複雜、未受控制的光線下仍能成功運作。

關鍵字

電腦視覺；光線變化；球形調和函數；人臉辨識；形狀重建

並列摘要

Lighting variation plays an important role on the appearance of objects, as the patterns of shading, specularities, and shadows. While the diverse ways in which our world can be illuminated make this world fascinating, the resulting variation in appearance is a major source of difficulty for many computer vision applications. Modeling the lighting variation may seem intractable at first glance due to its infinite number of degrees of freedom. Despite the daunting nature of this problem, some notable developments including the spherical harmonics are introduced. The spherical harmonics representation provides a coherent mathematical framework for modeling general lighting conditions. The reflected light field is considered as a convolution of the incident lighting and reflectance properties of object surfaces. In effect, a diffuse surface can be viewed as a low-pass filter acting on the incident illumination signal since a diffuse surface has a smooth appearance even if the lighting is complicated. Making these ideas formal in a signal processing framework, the theoretical results lead to many interesting insights and have practical implications of many computer vision problems, such as recognition, lighting estimation, and reconstruction. In this dissertation, we examine some traditional vision problems and propose novel methods based on the above techniques. Restricted assumptions that simplify or even ignore the lighting variation are relaxed such that the proposed methods work under general lighting conditions in uncontrolled environments.

並列關鍵字

computer vision ； lighting variation ； spherical harmonics ； face recognition ； shape reconstruction

參考文獻

for changes in illumination direction. IEEE Transactions on Pattern

[2] E.-W. Bai. An optimal two-stage identification algorithm for Hammerstein-

Wiener nonlinear systems. Automatica, 34(3):333–338, Mar. 1998.

[3] E.-W. Bai and Y. Liu. On the least squares solutions of a system of bilinear

equations. In IEEE Conference on Decision and Control, and European Control

國際替代計量

光線變化及其對視覺問題之應用

全文下載

主題瀏覽