結合PCA與灰色理論之人臉辨識系統

Abstract

本論文利用結合PCA與灰色理論來建立一套人臉辨識系統. 我們利用加入一些影像前處理,如用兩眼的間距來做正規化,用兩眼中心來做正臉位置的校準及利用統計正臉的平均灰度值來做平均灰度值的補償,來增加PCA理論的辨識率.另一方面,在第二階段辨識的方法是利用幾何特徵值來補償抽象特徵值PCA理論的不足.而幾何特徵值是利用影像處理常用的統計法,色調特性及邊緣偵測等技術將幾何特徵有效率及準確地求得. 利用計算第一階段用PCA理論所得結果的準確率將兩階段式不同辨識方法及特徵值結合.經過實驗過後可以很清楚的得到約93%的辨識率,而所花的時間在Pentium 700M Hz的電腦上平均不到1秒鐘.

This thesis proposes a human face recognition system combining PCA theory with Gray theory. Recognition rate of PCA theory can be improved by several image pre-processing, such as using distance between two eyes for normalization, using center of two eyes for adjustment and using gray balance. On the other hand, recognition of geometric features can be used to compensate recognition of abstract features using PCA. Several general image processings are used in the proposed system, such as histogram, color character and edge detection, to extract geometric features accurately and efficiently. Two recognition methods using distinct features are combined by calculating accuracy index of the result obtained from PCA. The proposed system obviously obtains about 93% recognition rate in several experiments. It spends less than one second in recognition on the Pentium 700M Hz computer.