- 學位論文
一個以半徑對稱性來加強人臉偵測可靠度的方法
A robust method using radial symmetry for reinforcing face detectable credibility
摘要
人臉偵測的技術在近幾年來已經相當的成熟，而其過去的眾多做法中，最常使用的方法是採用以膚色特徵來切割出臉孔範圍，再搭配人臉其它特徵的擷取配合使用，如檢查眼睛及嘴巴的特徵是否存在，以加強人臉偵測的可靠度。然而使用膚色來切割出人臉的方法存在許多問題，例如當背景的顏色與膚色相近時、光源的變化、以及不同種族(如黑人、白人、黃人)所呈現的膚色不同導致無法使用單一的色彩模組來處理等問題，都會造成誤判的結果，使得人臉偵測的正確率下降。 本論文提出以半徑對稱轉換模組來加強人臉偵測的可靠度。半徑對稱轉換模組是一個以梯度為基礎的運算元，用來偵測具有高對稱性的點。利用每一個像素點對在它周圍像素的對稱性所產生的貢獻度，以半徑對稱轉換模組來將具有對稱性的點變暗或變亮，可以有效解決前述各項問題；即利用人臉五官及臉孔與生俱來的對稱性，包括眼睛、鼻子、嘴巴等來加以確認，以更進一步確認是否為人臉，如此可大大的增加人臉偵測的可靠度，減少誤判的機會，並提出強化半徑對稱轉換模組於人臉偵測上的應用。
並列摘要
In the past few years, the need of detecting human faces plays an important role in many visual applications, and the amount of related technologies are growing rapidly. Among the technologies of human face recognition, the method by using the skin color of the face to segment the face scope is one of the most commonly used technologies. However, there are many problems that may exist when using the skin color to segment the face scope. For example, it may difficult to segment correctly the human face from the background when the background color is close to the skin color of the face. Another example is that the segmentation method using single color model may not work well when people of different race (for example black, Caucasian, and yellow person) are considered. In this thesis, we propose a robust method using radial symmetrical transformation module for reinforcing the credibility of face recognition. Radius symmetrical transformation module is a gradient-based operator that could be used to detect points of high radial symmetry. Since the facial features of human including eyes, noises, and mouths have the property of symmetry, the result of our method can reinforce greatly the credibility of face recognition and reduce the rate of erroneous judgment. The experiments are prepared to proof the feasibility and correctness of the proposed method.