影像處理在門禁系統應用之研究

Abstract

在本篇論文中我們整合了紅外線、數位攝影機、和人臉辨識的演算法，建構出一套門禁管制系統。在擷取特徵部份，我們先利用像差的概念框住整個人臉大小，由於人臉上較明顯的特徵為頭髮、眉毛和眼睛，我們利用二值化的影像，增強以上三部份的特徵，利用灰階特性定位出眼睛的位置，並以兩眼間之相對距離定位出整個臉的大小；在人臉辨識方面，利用Karhunen-Loeve expansion的方法找出每個人的特徵臉，並加以辨識。系統並提供了調閱檔案、搜尋出入時間資料庫、智慧型判別是否為人臉…等功能，使整個門禁管制系統更趨完美。在本篇論文更做了兩項實驗，一是比較以最小距離和最大內積和的特徵向量做辨識之差別；另一個是探討特徵擷取和辨識率之間的關係。整個實驗的目的是測試這套系統的功能，並比較不同辨識法則之差異性。

In this thesis, we have integrated infrared rays, digital camera, and algorithm of face recognition to build an entrance monitoring system. In capturing feature part, we first use the differential image guide to locate the whole head. Because the more clear features on the face are hair, eyebrows and eyes, we take advantage of binary image to enhance features of these parts. And then we use gray level property to find the location of eyes, and take the distance between the eyes to define the size of face. In face recognition part, we adopt the method of Karhunen-Loeve expansion to find eigenface for everyone and then to recognize it. The entrance monitoring system also support the ability of looking files, searching the entrance date and time in database, intelligently determining a face if a real face…etc. This can make our system more wonderful. In this thesis, we also made two experiments. One is to compare the difference of using the minimal distance to recognize and using the eigenvector of the maximal dot sum to recognize. The other is probing into the relation between the feature capturing and the percentage of recognition. The experiments purpose is testing the function of this system and comparing the difference of different recognition algorithm.