基于多尺度Gabor变换的人脸识别算法优化

Optimization of face recognition algorithm based on multi-scale Gabor transformation

针对人脸识别中的光照变化和姿态问题,提出首先获取已经过归一化处理的人脸图像(几何尺寸归一化和光照归一化),然后利用多尺度二维环形对称Gabor小波变换对归一化的人脸图像进行特征提取,再用局部二值模式对高维数据进行降维,最后使用直方图交的方法进行人脸相似度判决,从而实现较高识别率的实时人脸识别。实验表明,在FERET、Yale库和自建人脸库中的识别率得到了明显提高。

A method is proposed for illumination and pose changes of face recognition.Firstly,conducting the face image pre-processing included scale normalization and illumination normalization.Then,using multi-resolution CSGT to extract the normalization face image features,reducing high dimensional complex frequency domain magnitude response with local binary pattern is done latter.Finally,using histogram intersection for face similarity judgments to achieve higher recognition rate of the real-time face recognition.Experiments show that the face recognition rate has been improved significantly in the FERET,Yale and self-built face database.