机构地区: 中南大学信息科学与工程学院
出 处: 《计算机应用研究》 2009年第8期3152-3155,共4页
摘 要: 基于BCD提出了一种新的面部特征点定位方法,该方法在以下三个方面扩展了传统的BCD(boostedcascade detector):a)建立了BCD决策响应与特征点位置之间的概率关系;b)提出一种基于上述概率关系的特征点定位方法;c)设计了两种最佳人脸候选区域的选择方法。解析式的BCD把人脸检测和面部特征点定位融合成一个统一的过程。实验表明其精度和速度高于平均位置法(AVG)和基于boosted classifiers的最佳命中法(BestH it),并且它的运行速度也明显高于基于非线性优化的AAM和SOS法。 This paper described a novel technique called analytic boosted cascade detector (ABCD) to automatically locate features on the human face. ABCD extended the original boosted cascade detector (BCD) in three ways: a) a probabilistie model was included to connect the classifier responses with the facial features, b) formulated a features location method based on the probabilistic model, c) presented two selection criterions for face candidates. The new technique melted face detection and facial features location into a unified process. It outperformed average positions (AVG) and boosted classifiers + best response (BestHit). It also shows great speed superior to the methods based on nonlinear optimization, e.g. AAM and SOS.
领 域: [自动化与计算机技术] [自动化与计算机技术]