基于Haar特性的改进HOG的人脸特征提取算法

现有的大多数特征提取算法在提取人脸特征时,容易受到光照等外界因素的影响,从而导致后期人脸识别率下降。而方向梯度直方图(Histogram of Oriented Gradient,HOG)具有较强的光照鲁棒性,能够很好地减少由光照带来的干扰,但传统HOG在计算梯度幅值和方向时只计算水平和垂直方向上4个像素点对中间像素的影响,当外界环境变化时不能保证稳定性,因此提出一种基于Haar特性的改进HOG的人脸特征提取算法。该算法在计算梯度幅值和方向时考虑水平、垂直以及对角线上8个像素点对中间像素的影响,由于增加计算量导致特征提取时间也随之增加,因此引入Haar,借助Haar型特征运算简单、快捷的特点设计4组Haar型特征编码模式,按照改进的HOG特征计算方式提取人脸特征。在有光照等外界因素影响的FERET人脸数据库和Yale B扩展的人脸测试库中进行实验,实验结果表明,与GFC,LBP和其他文献中的HOG算法相比,该算法对光照具有更好的鲁棒性,能够在光照变化的环境下提高人脸识别率。该算法在FERET探测集fb,fc,dup1和dup2上的识别率分别为95.1%,80.9%,70.1%和63.2%,在Yale B中的识别率为89.1%。

Improved HOG Face Feature Extraction Algorithm Based on Haar Characteristics

Most of existing feature extraction algorithms are prone to be influenced by external factors such as illumination,which can lead to the decrease of face recognition rate.The robustness of histogram of oriented gradient (HOG) can solve the problem that brought by illumination on face recognition rate.However,when calculating the gradient direction and amplitude of pixels,the traditional HOG algorithm considers only the impact of the four pixels situated in horizontal and vertical direction.The gradient direction and amplitude of pixels may become unstable when the external environment changes.Thus,we proposed an improved HOG face feature extraction algorithm based on Haar characteris-tics.When calculating the gradient direction and amplitude,we considered the influence of 8 pixels.Meanwhile,because of the simple and fast operating of Haar-like features,we inducted Haar into HOG.We showed four groups of Haar feature encoding models,which calculated the texture features of face according to the improved HOG.In our experiments we used FERET and Yale B datasets.Experiments demonstrate that,compared with existing algorithms,the proposed method has better robustness and improve the recognition rate under varying illumination conditions.On the fb,fc,dup1 and dup2 datasets,the recognition rates of the proposed method are 95.1%,80.9%,70.1% and 63.2% respectively.On the Yale B datasets,its rate is 89.1%.