新聞室：成功大學論文研究色情圖片防堵色情網站

論文題目：基於膚色之裸體影像偵測之研究
研究生　：胡冠宇

下載：成功大學學位論文全文系統
http://etdncku.lib.ncku.edu.tw/ETD-db/ETD-search-c/view_etd?URN=etd-0730104-131644


摘要(中)
　　本論文在以膚色為基礎的裸體影像偵測研究上，提出一個新方法，以將裸體影像分成三大類特徵模組的方式來加以偵測。首先，採用Bayesian YCbCr膚色偵測模型，來做為膚色偵測的方法。再將含大量膚色的影像丟入裸體影像分類流程中，以分類出裸體影像和非裸體影像。而這些分類流程中，包含臉部特寫影像分類器，而用來過濾會影響裸體偵測效能的臉部特寫影像；全裸影像分類器，用來過濾膚色大量集中在軀體上，影像中人物一絲不掛之全裸圖；胸前特徵影像分類器，利用顏色和亮度上的差異，以及胸前特徵與膚色間之關係，來切割出胸前特徵，並以之作為裸露胸前特徵影像判別之依據；裸露私處部位影像分類器，利用快速物件分析的方式，分析膚色像素在影像上的分布，來切割出私處部位，並以之作為判斷之依據。本論文，測試了從網路下載總共1538張影像，研究成果有89.79%正確率，在未來不但有助於色情影像上的防堵，也能提供作為影像內容擷取技術上的參考。


摘要(英)
　　This thesis presents a new method for judging whether there are any naked people in an image based on skin color. The system consists of five stages. At first, The Bayesian YCbCr Skin Color Model is used to detect naked skin areas of the image and segmented it out roughly. Second. Haar face detector is used to avoid a facial closed-up shot. And then, a full-naked body image classifier is used to filter out the images which contain a naked people with no clothes on his body, and the nipple classifier is used to filter out these images which appear nipples on the human body. Finally, the private part classifier is used to filter out the images which contain people with her privates naked. Our proposal algorithm perform a recognition ratio of 89.79% with testing total 1538 images which were downloaded from the Internet and it is helpful both for the blue-picture-website-filter and for the semantic image indexing in the Content-based Image Retrieval.


關鍵字(中)
　裸體影像
　膚色偵測

關鍵字(英)
　skin color
　naked people image