基于非线性Boosting回归的多视角人脸配准

Non-linear Boosting Regression for Multi-view Face Alignment

人脸配准可以作为表情分析、人脸识别等人脸相关研究的预处理步骤,是人脸相关的计算机视觉研究的关键问题.本文针对图像中水平视角在正负45o内的人脸配准问题,利用基于Haar特征的非线性Boosting回归算法,根据标定点邻域内的局部纹理预测标定点的位移,提出了一种新的基于经典活动形状模型(Activeshape model,ASM)的实时多视角人脸配准算法.在两个数据集合上的测试实验表明,该算法在速度、准确度和稳定性上都比经典的ASM算法有显著提高且优于近期的改进算法,具有明显的实用价值.

Face alignment is a critical problem in many face related applications such as facial expression analysis, face recognition, etc. In this paper, we use local textures of each label point to predict the displacement of each label point by applying nonlinear boosting regression based on Haax rectangle feature, and develop a novel real-time multi-view face alignment system based on the active shape model. Experiments on two independent datasets show that our algorithm is much faster, more accurate and robust than the classic active shape model and outperforms recently improved algorithm, too. It has significant practical value in real applications.