利用二維與三維特徵之人臉辨識

Abstract

本論文的目的在於利用單一數位相機對被拍攝者進行連拍，並以畫面差異(Frame Difference)及邊緣偵測(Edge Detection)技術判定被拍攝者在影像中的位置，而後利用臉部色彩相似度的評估對臉部區域進行擷取，最後利用臉部區域的亮度資訊及臉部特徵的對稱性質，擷取影像中的臉部特徵點。當兩張影像的臉部特徵點對應已知，便可進行臉部特徵點的點位重建，並以兩組重建之後的臉部特徵點進行臉部特徵比對：利用旋轉、位移及比例縮放的轉換，將這兩組特徵點進行對齊，並根據臉部特徵點在3D空間的距離誤差，判定臉部特徵是否相符？

In this thesis a face reconginition based on 2D/3D facial features is addressed. A digital camera is set up to capture a video sequence of a rotating human face. First of all, a conventional frame difference technique is applied to detect the face region in each image. Then, the face region is further partitioned into skin and hair (non-skin) parts based on the color information. Facial features including corners of eyes and mouth, and ear contours are automatically extracted. After the correspondences of facial feature points are determinated, the 3D Euclidean facial feature points can be computed. Through the alignment of the two sets of the facial basis points in 3D space, the similarity between the two faces can be calculated according to the average distance between all the 2D/3D facial feature points.