摘要
针对人脸检测中对旋转、光照、遮挡等检测的准确性与稳健性问题,提出将连续均值变换用于人脸检测的方法。首先利用连续均值变换提取候选区域人脸的特征,用提取的特征训练SNoW分类器,通过分类器对人脸与非人脸样本进行分类,达到准确确定人脸位置的目的。实验结果证明,与人工神经网络、支持向量机和朴素贝叶斯相比,该方法在复杂背景和光照和遮挡以及多人脸等情况下仍然具有很好的准确性和稳健性。
To improve the problems of accuracy and robustness under the circumstances of rotate, covering and illumination in face detection, a method is presented which uses the Successive Mean Quantization Transform （SMQT）. Firstly, local face feature of candidate regional is extracted using SMQT. Then, the face feature is learned to train the SNoW classifier. Finally, the purpose of positioning faces is achieved accurately through classify face and face samples with the SNOW. Empirical results show the method outperforms than neural networks, support vector machines and Bayesian methods in accuracy and robustness even though with complex background, more than one faces, covering and illumination effects situations.
出处
《电视技术》
北大核心
2013年第1期154-156,共3页
Video Engineering
基金
河北省教育厅河北省高等学校自然科学研究指导计划项目(Z2010126)
