classification of human emotions remains an important and challenging task
for many computer vision algorithms, especially in the era of humanoid robots
which coexist with humans in their everyday life. Currently
本文提出了一种基于深度神经网络结构的鲁棒人脸对齐方法 Deep Alignment Network (DAN),DAN 采用全脸图像进行人脸对齐而非局部补丁,并通过在算法的每个阶段利用地标热图来提供之前阶段的视觉信息,从而使 DAN 处理具有大头姿态变化和困难初始化的面孔图像。我们在两个公开数据集上进行了广泛评估,结果表明 DAN 将失败率的最新技术水平降低了高达 70%。