可見光影像與熱影像之人臉辨識;Face Recognition Based on Visible Images and Thermal Images

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/72291

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/72291

Title:	可見光影像與熱影像之人臉辨識;Face Recognition Based on Visible Images and Thermal Images
Authors:	王麒堯;Wang,Chi-Yao
Contributors:	資訊工程學系
Keywords:	人臉辨識;生物特徵辨識;熱紅外線;face recognition;biometrics;thermal infrared
Date:	2016-08-30
Issue Date:	2016-10-13 14:37:28 (UTC+8)
Publisher:	國立中央大學
Abstract:	在現今的社會身分辨識愈來愈受到重視，在安全上扮演了相當重要的角色。其中最受矚目的是利用生物特徵作為辨識，不論是利用指紋辨識使用者的商品或是機場的海關人員透過虹膜辨識入境者的身分，都說明生物特徵辨識系統既便利又有可靠性。過去有許多關於生物辨識的研究¬¬──指紋、聲紋、掌紋、人臉和虹膜，利用各種演算法從中找出穩定又因人而異的特徵作身分辨識。本篇論文著重於人臉的辨識應用，所使用的人臉影像有可見光影像和人體溫度的熱影像，我們將結合兩者的資訊來做辨識。這兩種影像各有辨識應用上的優劣，而且捕捉的光波範圍不同，表示各有不同的資訊包含其中。可見光影像的部分使用人臉外貌的部分，透過經典的Fisherface方法取出特徵；熱影像的部分則是利用人體的生理現象，擷取皮膚溫度分布的特徵，透過溫度梯度和形態學找出一個類似血管分布的網路圖，我們使用局部正方形滑過網路圖，計算區域內的網路像素的數量作為特徵向量。最後結合這兩種特徵向量得到更長的特徵向量，再利用KNN分類器與資料庫的影像作比對、分類。實驗證明比起使用單一特徵用多個特徵作辨識效果更好。;Nowadays, human identification is more and more important in security. The most important identification method is the use of the biometric feature. Either the commodities which recognize the authorized user with fingerprint or customs officers use the iris recognition system to identify passengers, they elaborate the convenient and the reliable of biometric identification. In the past, a lot of researches on fingerprint, voiceprint, palmprint, human face and iris. They use kinds of algorithms to find out stable feature which differs from person to person for identification. In our approach, we devise a method combine visible images with thermal images for identification. These two kinds of different images have pros and cons. They capture electromagnetic radiation in different ranges and show they include different information. Features extracted from visual images by classical method, Fisherface method. From thermal images, we get the temperature distribution, by physiology phenomenon, based on temperature gradient and morphology. We use the local square windows to count pixels of a net to make feature vectors indicating images, which is called counter filter. Finally, we use two feature vectors and turn them into longer vectors, and then classify them with KNN classifier. Experimental results demonstrate that the performance of the system with multi-model is better than one with a single model.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	457	View/Open

社群 sharing

Loading...