多尺度变换域内混合投影熵的人脸特征描述

Facial feature descriptor using hybrid projection entropy in multi-scale transform domain

提出一种新的人脸特征描述方法。使用Contourlet变换提取人脸图像低频子带,并对子带图像适当分块从而减少图像局部扭曲对识别的影响,利用混合投影函数和图像熵提取特征从而构建混合投影特征矩阵。在ORL、Yale、CMU PIE人脸数据库的实验表明该方法具有一定的优势。

This paper proposes a new facial image feature description method.This method uses the Contourlet transform to get the low frequency sub-band and divides the sub-band image into several appropriate non-overlapping blocks so that the local image distortion will less affect the recognition result.It uses the hybrid projection function and image entropy to extract the features and construct the hybrid projection feature vector.The experimental results on ORL, Yale and CMU PIE face databases demonstrate that the new method is competitive.