雙眼彩色臉孔辨識之研究

Abstract

在這篇論文中我們提出一種利用立體視覺模型所做的彩色臉孔辨識方法。 首先，將兩個攝影機並排以達成如人眼般的視覺模型，並對攝影機前的人 取兩張影像。藉著由各個影像中找到特徵點並將兩張影像的相同特徵點做 對應，可以算出特徵點的三度空間位置。辨識所需的主要特徵值都是從這 些位置所導出的，另外還加上由區域邊緣點所導出的不變量。因為使用三 度空間位置，可以避免做大小的調整，也可以避免因為投影所造成的誤差 。本系統的訓練過程中含有特徵選取功能，可以選擇分離度較大的特徵用 於辨識。本系統的辨識方法是估計不明人物是資料庫中某人的機率。除此 之外，為了容易研究與使用，我們為本系統發展了一個方便的PC上的使用 者介面。 In this thesis, a color face recognition system using stereo vision modelis proposed. Two cameras are calibrated to be parallel and an image pair is captured from a person. 3D positions of feature points on face are foundby extracting their projected points on each image firstly and finding matches between two images. The feature vector includes distances and ratiovalues derived from feature points and invariant moments calculated from edge points. Normalization of positions of feature points and projectionerror can be avoided since 3D positions are used. The training processconsists of a feature selection algorithm that chooses features with higherdisjunction degree. The recognition process estimates the a posteriori probability of the unknown belonging to each person. A user friendly inter-face is developed on PC for convenient research and use.