本篇論文描述的是利用彩色資訊於以區域及特徵為基礎的立體視覺, 通常
在取得一組影像後, 最困難的問題就是影像的對應 , 大部份用來解決此
問題的方法, 可區分為以區域為基礎和以特徵為基礎的演算法。以特徵為
基礎的方法包含取出可能對應的階段和選擇唯一對應的兩個階段, 本篇論
文使用以區域為基礎的演算方式, 是將複雜物體的表面參數化成一連串的
三維區域, 利用最大概似估計法可以來估測這些物面參數, 經過一些推導
後, 可簡化成最小化一個誤差函數。傳統上對使用灰度層次的影像上某些
區域, 其亮度變化可能因相似而導致錯誤對應, 而彩度變化相似所發生的
頻率很小, 因此選擇使用彩色資訊來增加對應的正確性。我們引用彩色資
訊的方式, 是將一些彩色模式如RGB model , HSI model 等與誤差函數結
合成為新的誤差函數。本篇論文亦將描述和使用不同的彩色模式所得到的
效果。實驗結果告知我們在影像對應的表現上, 彩色資訊的使用有很大的
改進。此外本篇論文也提出一個實際的經驗, 利用立體視覺系統來引導機
械手臂做撞球實驗, 其中利用到以特徵為基礎的立體視覺 , 基本上這個
實驗可以分為三個部份, 視覺的校正, 手眼的校正和決定撞球的位置, 彩
色資訊對於找到不同顏色的撞球和決定它們的位置有很大的幫助。

This thesis describes the use of chromatic information in
stereo vision. Generally speaking, the most difficult process
in stereo vision is stereo correspondence. Most existing stereo
correspondence algorithm may be classified as either feature-
based or region-based. The feature-based algorithms are
comprised of the matching extraction stage and the disparity
selection stage. The thesis uses a region-based stereo
technique which models surface of complex objects by a number
of 3D parameterized surface patches. Estimation of surface
parameters is then formulated as a maximum likelihood
estimation problem. Notice that gray-level intensity variation
is more likely to be similar in some regions, which would
result in false correspondence due to small noise. Since
similar chromatic variation occurs much less frequently, we use
chromatic information to obtain better accuracy in stereo
reconstruction. The effect of using different color models in
matching, such as trichromatic model , HSI model ......,etc,
are investigated. Experimental results show an obvious
improvement in the performance of algorithm when chromatic
information is used. Also, this thesis performs a real
experiment that use a color stereo vision system to guide the
robot for playing billiards, where the feature-based stereo
method is used. The chromatic information can be used to
segment the color billiards and helps in finding the stereo
correspondence of different billiards.