DETECTION OF FEATURE POINTS FROM LIP IMAGE SEQUENCES USING A TWO-STAGE SVM CIASSIFIER (Computer Vision, Medical Applications and Networked MM)(International Workshop On Advanced Image Technology (IWAIT2004))

概要

論文の詳細を見る
We propose herein an algorithm by which to detect four lip feature points from an image sequence. The feature points consist of the left and right side points of the mouth as well as the center points of both the upper and lower lips. It is possible to detect these points from an image which contains expected changes in both lighting conditions and the direction of the face. In the algorithm, the SVM classifiers are arranged for each feature point in two resolutions. First, in the low-resolution image, the rough positions of each feature point is detected using the global pattern information and the geometric placement constraint of these points. Then, in the high-resolution image, feature points are detected from the area decided by rough positions at first search. This detection is carried out based on the local pattern information. The two-stage search on resolution and a geometric placement constraint can improve the accuracy and computational cost of the detection.
社団法人電子情報通信学会の論文
2004-01-06