眼鏡・髭によらない顔特徴点の検出(一般セッション(4))(データマイニングとパターン認識・メディア理解)

概要

論文の詳細を見る
This paper proposes a new approach for detecting facial organs and extracting facial feature points from an input color image, which is not influenced by disguise, e.g., wearing glasses, with beard. Using both color information (skin color and hair color), and edge information (edge strength and edge orientation), we detect the minimum region (hereafter exact face region) that contains all facial organs by applying the integral projection method. And then we check whether the person is wearing glasses or not, by calculating two ratios of the information in the cheek area and the outside area of eyes. Judging from this result, facial organs can be extracted by 2 different algorithms. In order to acquire a stable and exact extraction result, we predict the organ's positions from the information of both the exact face region and the already extracted facial organs. Finally, we extract the facial feature points from each facial organ by using the SUSAN corner detector. The extensive experiments over many real images demonstrate the robustness of our method.
社団法人電子情報通信学会の論文
2003-09-02