基于对称双线性模型的光照鲁棒性人脸表情识别

针对传统的光照预处理方法降低原始图像质量、丢失部分有效辨识信息的缺点,提出一种新颖的应用对称双线性模型来对人脸表情图像进行光照预处理的光照鲁棒性人脸表情识别方法.首先通过对称双线性模型将训练集图像分解为相互独立的光照因子和表情因子,并提取其光照因子.接下来提取含有未知光照的测试集表情图像的表情因子,并将其转换到训练集的若干个已知光照上,这样处理能够将任意光照的测试图像转换到相同的光照平台上,令所有测试图像的特征具有归一化特性.实验结果表明, 本文所提光照预处理方 法在识别性能上优于传统的光照预处理方法,应用在光照处理后的JAFFE表情库上识别率达到92.37%, 表明其适用于光照鲁棒性人脸表情识别.

Illumination-robust Facial Expression Recognition Based on Symmetric Bilinear Model

This paper proposes a novel illumination-robust facial expression recognition method using symmetric bilinear model to overcome the disadvantage of traditional illumination preprocessing methods that can degrade the quality of input image and worsen recognition performance. Firstly, the illumination factors are separated from the training database. Secondly, the expression factor is separated from testing image with arbitrary illumination. Thirdly, the testing image is transformed into a number of expression images exhibiting different illuminations of training database. Experimental results show that the proposed method is better than the traditional illumination preprocessing methods in recognition performance. The best recognition rate of 92.37% is achieved in JAFFE database with illumination transformation, indicating this method is suitable for illumination-robust facial expression recognition.