基于Curvelet变换特征的人脸识别算法

FACE RECOGNITION ALGORITHM BASED ON CURVELET TRANSFORM FEATURE

针对小波变换无法准确表达二维奇异曲线的弱点,提出基于曲波(Curvelet)变换特征的人脸识别算法。Curvelet变换可以很好地去逼近奇异曲线,对于人脸图像能实现最优的稀疏表示。该算法采用基于Wrapping的离散Curvelet变换加权算法对训练集的人脸图像进行特征提取生成特征矩阵,再通过PCA方法降低维数后结合稀疏表示分类算法(SRC)进行人脸识别。通过在ORL、Yale和AR三个人脸数据库上的仿真实验以及和基于小波变换类识别算法、LDA算法和SRC算法等比较,实验结果表明该算法在人脸遮挡、姿态变换、表情变换和光照变换等干扰因素的作用下具有较高的人脸识别率和较好的鲁棒性。

In view of the weakness that the wavelet transform cannot express the two-dimensional singular curve accurately, a face recognition algorithm based on Curvelet transformation was proposed in this paper. Curvelet transform can be very good to approximate the singular curve, for the face image can achieve the optimal sparse representation. In this paper, we used the Wrapping-based discrete Curvelet transform weighting algorithm to extract feature matrices from the face images of the training set, and then used the PCA method to reduce the dimension and combined the sparse representation classification algorithm （SRC） for face recognition. Through the simulation experiments on ORL, Yale and AR face database and the comparison with wavelet transform class recognition algorithm, LDA algorithm and SRC algorithm, the experimental results showed that the proposed algorithm had high face recognition rate and good robustness under the influence of interference factors such as face occlusion, gesture transformation, expression transformation and illumination transformation.