信息熵加权的HOG特征提取算法研究

Research on HOG Feature Extraction Algorithm Weighted by Information Entropy

针对人脸图像中不同部位所含的信息熵不同,对识别的影响程度不同等因素,提出了一种信息熵加权的HOG特征提取方法。该算法将待识别的人脸图像进行分块,对分块后的图像进行HOG特征提取,计算每块图像所含的信息熵作为权重系数加到各个分块中形成新的HOG特征,通过PCA算法对特征进行降维,得到信息熵加权的HOG特征。通过在ORL和YALE实验结果表明,该算法相较于其他传统识别方法具有更高的识别精度和准确度,并且对于人脸在光照、姿态表情等干扰因素下均具有良好的有效性和鲁棒性。

According to the different information entropy in different parts of face image,the influence of different factors on the recognition degree are different,this paper proposes an information entropy weighted HOG feature extraction method.The facial image to be identified is divided into blocks,HOG feature extraction on the block of the image,and then this paper calculates the information entropy of each image contained as weight coefficient to each block in the formation of new HOG features,the features are reduced by PCA algorithm,and the HOG features of information entropy weighting are obtained.The contrast experiment on ORL and YALE shows that this method not only has higher recognition accuracy than other traditional recognition methods,but also has good robustness and effectiveness for transforms of illumination,face pose and expression.