- 學位論文
即時頭髮輔助之人臉辨識
Real-Time Face Recognition in aid of Hair Detection
摘要
在本篇論文中，我們提出一個新的即時人臉辨識方法，在頭髮辨識的輔助之下人臉辨識率會因為使用了我們的合併演算法而提高。 在本篇論文中，我們提出了一個新的人臉辨識系統，系統分為兩個部分。第一個部分為臉部的抓取與辨識，第二個部分則是頭髮的抓取與辨識並且使用合併演算法與第一部分做結合。在第一部分為了找出膚色區塊，我們先把RGB的圖片轉換成YCbCr之後再轉成二值化圖片，並使用Cb與Cr來當做臉部辨識的色彩空間，我們會使用預先設定好的膚色閥值範圍來找出膚色的區域，之後在使用膨脹侵蝕以及四連通消除多餘的雜訊，之後我們在使用互相關的方法將橢圓形的圖片拿去與整張圖做重疊搜尋，如果其中的膚色區塊符合橢圓形的形狀就會被框選出來。在第二部分我們則是使用Y的色彩空間來做頭髮的辨識，流程跟臉部一樣是消除雜訊與使用互相關來抓出頭髮，然後利用頭髮較好的辨識率提升人臉的辨識率。最後則是會介紹我們的模擬與實驗，和與其他論文的比較。 我們的貢獻如下所示: (1)提出了一個新的人臉辨識方法: 我們使用的人臉辨識系統與其他人所使用的方法不同，我們的演算法與他人比起來較簡單與直觀，我們使用了頭髮的部分來改善臉部的辨識率。 (2)提升圖片處理的速度: 因為我們使用的是區塊的辨識方式，所以我們可以利用著個特性來快速的移除大面積沒有意義的雜訊，這會讓框選人臉的時候更加的快速與準確。
並列摘要
In this thesis, we will propose a new real-time way to recognize human face. With the aid of hair recognition the correct rate of face recognition would be able to increase by our algorithm. In this thesis, we propose a new face recognition system. The system have two parts. One is face capture and recognition. The other part is hair capture and recognition which is used to be combined to the first part. On the first part, in order to find out the skin color, we transform the image into binary image by converting it from RGB to YCbCr. We use the preset threshold to find out the skin region. After that, four connected erosion and dilation is used to reduce the noise. Then we choose a bigger area to be the recognized skin color region. After that we use human face average proportion which is defined in statistics to limit the width and length to locate the face region. On the second part, we distinguish hair region from background by using Y space. After that the cross-correlation is used to find out the hair part with preparatory proportion image. The hair region can be adapted to be combined to the first part to help the recognition of face region. Finally, we will present the simulation experiments and the comparison with the other researches. The contribution of our research are as follows: Provide a new algorithm of face recognition: Unlike previous face recognition systems, our algorithm is easier and more intuitive by using hair part to improve the correct rate of recognition. Increase the speed of image processing: Because the block identification is used by us, we can remove no significance region abundantly and quickly. It will locate the face region more rapidly and precisely.