一种基于小波分量变换的人脸图像光照归一化算法

基于小波变换原理,提出一种基于小波分量变换的人脸图像光照归一化算法.人脸图像经过二维离散小波变换(DWT),被分解成4个子分量(LL,HL,LH,HH).将低频分量(LL)进行对数变换和分段线性变换,对高频分量(LH,HL,HH)进行Gamma变换.对所有子分量进行小波逆变换,对经小波重构后的人脸图像进行中值滤波.分别在Yale B和CMU-PIE人脸数据库中对本文算法进行光照归一化有效性试验;对比本文算法与其他22种光照归一化算法的处理时间及处理效果;进行分段线性变换和伽马变换参数比较试验及人脸识别试验.结果表明:本文算法执行速度快,处理效果好,人脸识别率高,适用于不同光照条件的人脸识别系统.

Illumination Normalization Algorithm for Face Image Based on Wavelet Component Transform

A wavelet transform based face image illumination algorithm is proposed based on the principle of wavelet transform. A face image is decomposed into four sub components by two-dimensional discrete wavelet transform (DWT),which are low frequency components (LL) and high frequency sub components (HL,LH, HH). The LL component is transformed by logarithm and piecewise linear transformation,and the LH,HL and HH components are transformed by Gamma. All sub components are used to wavelet reconstruct ( IDWT) and median filter the face image. In Yale B and CMU-PIE face database,experimental effectiveness of light treatment of the algorithm, the processing time of this algorithm compared with 22 other illumination normalization algorithms and the treatment effect were carried out, parameter piecewise linear transformation of comparative experimental comparison test and gamma transform parameters, face recognition experiments were carried out successively. The experimental results show that the proposed algorithm is fast,effective,and has high recognition rate. It can be applied to face recognition system with different illumination conditions.