摘要
人脸图像的预处理具有重要意义,是人脸系统走向实用化的关键一步。精度、速度和鲁棒性是算法的3个关键问题。这里的算法利用肤色的先验知识缩小搜索区域,Adaboost检测后准确定位人脸;将Sobel边缘信息结合灰度初步确定眼睛位置和眼球半径;使用眼睛模版与眼球下半边缘轮廓匹配,最终精确定位眼睛并完成标准化。算法基本满足实际系统对上述3个关键问题的要求。
Image pre- processing plays an important role in making face recognition system practical. The accuracy, rate and robost of processing are three key criterions to evaluate the algorithm. The method we proposed firstly extracts the skin color regions,then precisely detects the human face using the Adaboost algorithm. Sobel edges are combined with their gray image to get the eye location and the eyeball radius. After that, the eye templates are matched with the contours of underside eyeballs,the pupil centers are extracted precisely, and the face image is normalized finally. The simulation results show that this pre - processing algorithm efficiently make trade off among the three key criterions. face
出处
《现代电子技术》
2007年第24期115-118,共4页
Modern Electronics Technique