人臉辨識之研究

Abstract

本論文提出一套人臉辨識系統，系統主要分為臉部的定位切割與辨識兩部 份。在定位及切割部份，本論文提出一個自動的定位及切割方法，它主要 是根據人臉部五官的對稱性、頭髮與眼睛的相關位置以及它們在影像中的 灰階值特性，找出眼睛的位置並進而切割出臉部的區域。我們將這些所切 割出來的臉部影像的大小標準化，以供辨識之用。辨識部份主要是採用特 徵臉辨識法，由於切割出來的影像只包含臉部，因此可以免除背景對於特 徵臉方法的影響，而為了降低光線對特徵臉方法的影響程度，我們將所有 的臉部影像作一些前處理的動作。根據實驗結果，所提出的定位切割方法 在不同的光線、背景及距離等因素下都有很高的正確率，證明此方法的可 行性。而在辨識部份，將臉部影像作前處理的動作，明顯地提高了特徵臉 方法的辨識率。論文中並提出一個人臉辨識實際使用於門禁管制的架構流 程，可以作為日後實作一個系統的藍圖。 An automatic face recognition system is proposed in this thesis. The system mainly consists of two parts : location and recognition. In location part, a face location algorithm is proposed. The symmetry of a face and the gray level characteristics of hair and eyes are used clip the face region out of an image. The face images clipped in the location phase are normalized to the same size and are used in the recognition phase. The eigenface approach is used for recognition. Since the eigenface approach is very sensitive to the lighting condition, some preprocesses to the face images are used to reduce the influence of illumination. Experimental results show that the proposed face location algorithm is robust under the changes of illumination, background and distance. The preprocesses to the images improve the recognition rate of eigenface approach. An architecture of an entrance control system using face recognition are proposed and can be used as a basis to build a real system.