Abstract: To solve the deficiency of describing face information by single feature, a novel method of face detection based on multiple facial feature extraction is proposed. Wavelet analysis is used to reduce the dimensions of the candidate face getting from skin segmentation, and discrete cosine transform is used to extract the coefficients as frequency domain features. Both the algebra features and the texture features are extracted from discrete cosine transform reconstruction image by singular value decomposition and local binary mode respectively. These extracted features are combined to be a new characteristic vector with much lower dimensions than the original image, while its ability to characterize is still strong because of the comprehensive advantages of three features. Support vector machine is applied to classifying and locating human face. The experimental results show that this method has good detection rate and high robustness.

Key words: face detection, discrete cosine transform, singular value decomposition, local binary mode

摘要： 针对单一特征在人脸检测方面的不足，提出了一种基于多特征提取的人脸检测算法。利用肤色信息分割出候选人脸区域，并对其进行小波分析，降低维数。进行离散余弦变换，取出部分系数作为频率域特征。对变换后的重构图像利用奇异值分解和局部二值模式提取代数特征和纹理特征，将这三方面特征融合成新的特征向量。这样既降低了维数，又综合了三方面的特征优势，保证了利用支持向量机分类，定位人脸的效果。实验结果表明，该方法具有较高的检测率，且鲁棒性较好。

关键词: 人脸检测, 离散余弦变换, 奇异值分解, 局部二值模式