Abstract: A novel face detection method based on spectral histogram constructed by Dual Tree-Complex Wavelet Transform （DT-CWT） and Support Vector Machine（SVM） is presented in this paper.Face images can be properly represented according to good directional selectivity on different scales of DT-CWT.Then spectral histograms are constructed by DT-CWT and other filters convolving with original images.Spectral histograms have intrinsic generalization of image representations.An efficient and robust classification function is obtained by training SVM using histograms as input vectors.Experimental results show that DT-CWT out-performed Gabor wavelet transform especially on computational redundancy and speed.And DT-CWT applies in face detection gives the satisfying detection accuracy.

Key words: Dual Tree-Complex Wavelet Transform（DT-CWT）, spectral histogram, Support Vector Machine（SVM）, face detection

摘要： 提出了一种利用双树-复小波变换（DT-CWT）构成频谱直方图并通过支持向量机（SVM）进行人脸检测的方法。在人脸图像的表示中，基于DT-CWT在不同尺度上具有的很好方向选择性，对原始图像滤波，并与其它滤波器卷积滤波后得到原始图像的不同频率特征一起构成频谱直方图，该直方图在图像的表示上具有很好的本质扩展性。通过支持向量机（SVM）对频谱直放图向量进行分类训练，得到了有效区分人脸与非人脸的分类函数。实验显示，DT-CWT具有与Gabor变换类似的性质，而且计算冗余度更小，计算速度更快。应用DT-CWT频谱直方图的人脸检测算法取得了令人满意的结果。

关键词: 双树-复小波变换（DT-CWT）, 频谱直方图, 支持向量机（SVM）, 人脸检测