利用判决回归的快速人脸特征定位算法  

Fast human face landmarks locating algorithm based on discriminative regression

作　　者：刘畅[1] 毕笃彦[1] 熊磊[1] 李岳云[1] 

机构地区：[1]空军工程大学航空航天工程学院,陕西西安710038

出　　处：《计算机工程与设计》2015年第9期2472-2475,共4页Computer Engineering and Design

摘　　要：提出一种在受限局部模型(constrained local models,CLMs)框架下基于判别回归的人脸表示方法,其对于非标记的人脸图像表述效果显著。该方法与树状模型相比,用很少参数表示响应图,高效地对未知的响应图进行重构,通过对其中回归方法进行优化,实现鲁棒的形状参数更新,提高运算速度。在Multi-PIE数据库上的实验结果表明,该算法的计算效率优于其它主流算法,具有实时处理的能力。A discriminative regression method was presented in constrained local models framework which showed a remarkable performance of representing unlabelled human face images. Compared with tree-based model, the proposed algorithm represents response map with a small set of parameters and reconstructs unseen images efficiently. In addition, the regression method was optimized to achieve robust shape parameter update efficiently. Results of experiments in Multi-PIE database demonstrate that the proposed algorithm outperforms other state-of-the-art methods and realizes real-time processing.

关 键 词：人脸表示 受限局部模型(CLM) 判别回归 实时处理 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]