人臉包含五官、毛髮等許多器官所組成的,是一個極為複雜的測試目標,不同的臉型的大小、五官的位置、皮膚的顏色與輪廓的深淺都有不同的變化性,為了要實現偵測人臉的正確性,必須要考慮人臉在各種複雜的環境中,取得最精確的數據,一般取得的影像通常為數位相機所拍攝或是網路攝影機所得到的影像,所以在光源的方向、明暗、色彩都會增加原影像的雜訊,造成影像處理的誤差也降低人臉偵測的正確率。 人臉的偵測是有一定的困難和複雜度,對於這一問題國內已有許多相關的研究與實驗,目前電腦視覺與人工智慧的發展也越來越成熟,國內外的研究也提出許多方法,但對於一種能夠普遍且適用於各種複雜環境的人臉偵測方法目前還有很大的探討空間。 人臉辨識系統可以分為兩個主要部分:人臉偵測與人臉辨識的方法。本篇論文是以膚色的分割與新穎的幾何方法,對人臉特徵的偵測方法做探討。 本論文結合膚色區域分割、邊緣偵測、水平垂直投影技術實現一套人臉偵測與五官的定位系統。此系統主要分為兩部份:人臉偵測與五官特徵定位。在人臉偵測部份,先將原始影像做正規化,再將顏色分割,找出膚色的範圍,對膚色影像做二值化,偵測影像中人臉所在的區域。在五官特徵定位的部份,將已找到膚色區域的範圍框取出來,運用影像邊緣偵測運算的方法找到五官特徵的輪廓,再利用水平及垂直投影的方式,找出眼睛所在的區域位置,並以距離計算眼睛、鼻子與嘴巴的位置,以定位出五官其特徵點的位置中心點。
Face features, include face, hair and organs, are composed of complicated human factors. Face detection has problems of different image size, location of facial features, skin color etc. In order to achieve the accuracy of face detection, it is necessary to consider the human face in a variety of complex environmental to obtain the most accurate data. Along with the light direction, brightness, shade and color would add to the original image noise, resulting decreasing the face detection rate in image processing. Face recognition system can be divided into two main parts: face detection and face recognition. This paper is based on novel geometric method for facial feature detection methods. Our approach combines color skin segmentation, edge detection (Sobel) technology to achieve a set of face detection and facial features positioning system. This system is mainly divided into face detection and facial features positioning. For the face detection part, the first process is the normalization for the original image,and then the color segmentation is used to identify the scope of skin color and mouth of the images. After that, mouth binarization takes place. the detection of human face is almost down. Facial features positioning is to find the scope of color of the input image. The edge detection skill uses the characteristics of the profile to find features, and then we use the horizontal to identify the location of the eye region, and Euclidean distance calculation of the eyes, nose and mouth position are used to locate the feature points.