基于多尺度分析矩特征的人脸表情识别  

Facial Expression Recognition Based on Multi-scale Analysis and Invariant Moments

作　　者：支瑞聪[1] 阮秋琦[1] 

机构地区：[1]北京交通大学信息科学研究所,北京100044

出　　处：《信号处理》2009年第5期692-696,共5页Journal of Signal Processing

基　　金：国家"九七三"重点基础研究发展规划项目(No.2004CB318005);国家自然科学基金项目(No.60672062;No.60472033);北京交通大学优秀博士生科技创新基金项目(48036)资助。

摘　　要：Zernike不变矩具有对噪声不敏感,正交等特性,是表情的有效表征方法,高阶Zernike矩包含更多图像信息,对表情分类的作用更大。但是高阶矩的计算复杂度很大,很难达到快速表情识别的要求。本文利用小波变换对表情图像进行多尺度分析,从低频子图像中计算其Zernike矩作为判别特征进行表情识别。通过小波变换,一方面可以对图像降维,降低计算复杂度;另一方面,小波变换的去噪性能使得识别效果更好。实验表明,基于多尺度分析Zernike矩特征的方法优于单独使用小波变换或Zernike矩特征方法的识别效果。Zernike invariant moments are orthogonal and insensitive to noise,and they are efficient for facial expression represen- tation, especially the moments with high order. Although high order moments are more useful to classification, the computation is com- plex. Therefore, we present a novel feature extraction method based on wavelet transform and Zernike moments. The facial images are de- composed using wavelet transform and Zernike moments of the low frequency domain are treated as feature vectors for facial expression recognition. Wavelet transform is used to reduce the dimensionality of the facial images and weaken the noise, so that the moments can be computed easily, and higher recognition accuracies are achieved. Experiment results indicate that this method outperforms the methods which use wavelet transform or Zernike moments alone.

关 键 词：ZERNIKE矩 矩不变量 小波变换 表情识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]