以膚色分割及類神經網路為基礎之人臉偵測

機構典藏 > College of Engineering > Graduate Institute & Department of Computer Science and Information Engineering > Thesis > Item 987654321/35012

Please use this identifier to cite or link to this item: https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/35012

Title:	以膚色分割及類神經網路為基礎之人臉偵測
Other Titles:	Face detection based on skin color segmentation and neural network
Authors:	王淑儀;Wang, Shu-yi
Contributors:	淡江大學資訊工程學系碩士班林慧珍;Lin, Hwei-jen
Keywords:	人臉偵測;膚色分割;倒傳遞神經網路;face detection;skin color segmentation;back-propagation neural network
Date:	2005
Issue Date:	2010-01-11 05:53:56 (UTC+8)
Abstract:	人臉偵測是一項具挑戰性的工作，也是人臉追蹤與辨識系統之重要的前處理部份，本論文提出一套人臉偵測的方法。首先，利用膚色資訊快速地找出人臉可能存在之區域，並得到一膚色二值化圖(Skin Map)，再對此膚色二值化圖做雜訊去除的處理以及型態學的運算（浸蝕、擴張與連通元件），並利用臉部的長寬比特性過濾出可能的人臉區塊。接著，對這些可能的人臉區塊做眼睛的偵測，若找到眼睛，則利用眼睛的位置預測人臉大小並框出可能的人臉範圍，即候選臉區塊，若找不到眼睛,則視為非人臉區塊。最後，利用類神經網路，針對候選臉區塊進行臉部驗證的工作。實驗結果顯示，本文可以有效地偵測影像中的人臉，並可以克服影像中人臉不同亮度、大小、旋轉與多人人臉的問題。 This paper proposes a human face detection system based on skin color segmentation and neural networks. The system consists of several stages. First, the system searches for the regions where faces might exist by using skin color information and forms a so-called skin map. After performing noise removal and some morphological operations on the skin map, it utilizes the aspect ratio of a face to find out possible face blocks, and then eye detection is carried out within each possible face block. If an eye pair is detected in a possible face block, a region is cropped according to the location of the two eyes, which is called a face candidate; otherwise it is regarded as a non-face block. Finally, each of the face candidates is verified by a 3-layer back-propagation neural network. Experimental results show that the proposed system results in better performance than the other methods, in terms of correct detection rate and capacity of coping with the problems of lighting, scaling, rotation, and multiple faces.
Appears in Collections:	[Graduate Institute & Department of Computer Science and Information Engineering] Thesis

Files in This Item:

File	Size	Format
	0Kb	Unknown	435	View/Open

Loading...