扩展型HMM在表情识别中的应用研究

Extended Hidden Markov Model for Recognition of Facial Expression

根据表情与人脸表情特征关系,提出采用隐马尔可夫模型进行人脸表情识别;又鉴于人脸图像二维特性,提出了更具健壮性、更易处理二维数据的扩展型隐马尔可夫模型.该模型相比伪二维隐马尔可夫模型,简化了复杂度.为提高模型的识别效率,根据敏感度不一,提出多重感兴趣区域替代单一的感兴趣区域.为提高表情子库内样本的聚合度及库间样本离散度,提出相应的改进方案.首先通过人脸检测,实现表情样本采集;然后采用二维离散余弦实现图像频域转化,并结合低频数据生成特征向量;最后采用扩展型隐马尔可夫模型进行表情建模,实现表情训练与识别.实验表明:采用扩展型隐马尔可夫模型可有效识别表情,尤其是优化后的设计方案.

The paper introduces an approach of facial expression recognition using Extended Hidden Markov Model （E-HMM） on the basis of the relation between facial expression and facial features. The proposed method can better model 2D-data than 1D-HMM with less computational complexity in comparison with the pseudo 2D-HMM. The proposed method is characterized in using multiple region of interest （multi-ROI） instead of single region and combining both spatial and temporal features with the E-HMM. The method makes use of an optimized set of observation vectors obtained from the 2D-DCT coefficients of the facial region of interest. The E-HMM is trained using segmental K-means algorithm and then used for the facial expression recognition. The experimental results reveal the significant system performance improvement and robustness as well.