請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/37059
標題: | Wii遊樂器之Mii頻道人像自動產生研究 Automatic Generation of Cartoon Faces for Wii Mii Channel |
作者: | Yen-Nung Lee 李硯農 |
指導教授: | 張瑞益 |
關鍵字: | 人臉識別,圖像產生,圖形比對,SVM分類器,Wii,Mii, Face Recognition,Portrait Generation,Pattern Matching,SVM Classification,Wii,Mii, |
出版年 : | 2008 |
學位: | 碩士 |
摘要: | Wii是目前當紅的遊樂器之一,而Mii頻道卡通肖像產生系統正為Wii受到矚目的重要原因之一。然而一般在使用Mii製作人像的時候,必須一個一個元件去選取,過程相當繁瑣及耗時,因此,我們想到如果可以結合影像處理的技術,整合人臉識別及圖形比對的流程,將人臉圖像經過識別之後,自動產生與原圖相似的Mii肖像,應該會是一個有趣且實用的研究議題。本論文將整個人臉產生的過程分為兩個部分,在影像前處理部分,首先對輸入的大頭照影像利用膚色空間轉換作擷取,在人臉範圍內針對不同的特性分別找出眉毛、眼睛、鼻子及嘴巴等元件,接著再利用彩色影像二值化、垂直與水平投影及邊緣偵測等影像處理演算法,標定每個人臉元件的特徵點。接下來將標記好的特徵點資料運算得到每個元件之特徵值,配合經由人臉資料集訓練過之支援向量機(Support Vector Machine)分類器,依據每個元件的特徵值找出該元件所屬類別,也就是與原始人臉影像最相似的Mii元件圖案,最後輸出完整的Mii肖像,完成整個系統。本研究實際開發出一個人臉圖像自動產生出Mii卡通圖形的系統,並以多張實際人臉影像作測試。實驗結果顯示,所提出的系統能夠有效根據真實人臉影像中各元件的特徵值,透過SVM分類器的選擇,找出與原圖相似的Mii圖案。 Wii is one of the most popular TV game consoles at present. It contains a fabulous cartoon portrait system called “Mii”. When we use Mii to produce a cartoon portrait, we need to choose all facial features (such as eyes, nose, mouth, et al.) step by step. This procedure is complicated and may not achieve an acceptable result even spending lots of time. Thus, we think it will be an interest and practical topic if we can utilize face recognition and pattern matching techniques to generate the related Mii portrait from a given photo. We separate the main procedure into two parts. In the pre-processing part, we process the input image to find facial features such as eyes, eyebrows, nose, and mouth. Then we use image process algorithms to locate the feature points on each facial feature. We calculate feature values by using these feature points, and input them to the SVM classifier, which has already trained by many portrait/photo samples, to classify what kind of Mii portrait is the most similar to the original photo. In this thesis, we have developed a system that can generate the related Mii portrait from a given photo automatically. Experimental results show that our proposed system can availably find a Mii portrait which is similar to the original photo. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/37059 |
全文授權: | 有償授權 |
顯示於系所單位: | 工程科學及海洋工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-97-1.pdf 目前未授權公開取用 | 2.78 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。