一种具有噪声鲁棒性的人脸表情识别算法研究

Research on Noise Robust Facial Expression Recognition

为降低噪声对人脸表情识别的影响,首先提出具有人眼视觉特性的各向异性扩散滤波方法,对图像进行滤波预处理;同时采用改进HOG算子提取人脸表情特征。实验结果表明,改进的各向异性扩散滤波算法在滤除噪声的同时能更好地保留表情图像的弱小细节信息,改进HOG算子相比传统特征提取算子可以更准确地描述人脸表情特征。因此,该算法是一种有效的、具有一定噪声鲁棒性的人脸表情识别算法。

In order to reduce the interference of noise with facial expression recognition. Firstly, a new anisotropic diffusion algo- rithm based on human visual system is proposed in this paper. Meanwhile, the paper makes improvements on HOG algorithm, and uses the advanced method to extract facial expression feature. The experimental results show that the improved anisotropic diffusion algorithm can preserve small details much better while filtering out noise and the improved HOG algorithm can describe the facial expression much more accurately compared with some other traditional descriptors. Therefore the proposed algorithm is an effective facial expression recognition algorithm with strong noise robustness.