基于曲波与主分量分析的人脸识别

本文提出了基于曲波变换和主分量分析的人脸识别算法。针对小波变换仅能有效表达图像中的点奇异性的弱点，采用曲波变换提取面部主要特征。由于人脸的主要特征是面部的曲线信息，而曲波变换直接以曲线为表达基元，其变换系数能有效表示沿曲线的奇异性，而且是各向异性的，所以能更好地表达面部特征。进一步使用主分量分析将特征投影到更具表达力的空间中，从而达到更高的识别率。实验结果表明曲波的性能优于小波，尤其是曲波小尺度系数的识别率明显高于小波高频系数。

Face recognition based on curvelet and PCA

A new method combining curvelet transform and principal components analysis (PCA) is presented for face recognition. Considering the disadvantage of wavelet, say, it is only optimal in representing point singularities, we use curvelet transform to extract facial features. The facial features being mostly curves, curvelet transform directly takes edges for representation, resulting in a more powerful feature extraction. PCA is then used to map the feature into more meaningful subspace, hence we get higher recognition rate. The experiments demonstrate the effectiveness of the proposed method.