分块Gabor结合梯度直方图的特征提取算法

Feature Extraction Algorithm Based on Block Gabor Combined with Histogram of Oriented Gradients

针对单一HOG特征提取方式会丢失图像局部特征的问题,本文提出了一种基于分块Gabor的梯度直方图特征提取算法.该方法首先将待识别的人脸图像通过Gabor特征提取方法得到图不同尺度和方向的图像Gabor特征,然后对Gabor特征按照尺度和方向两方面进行融合,之后再对融合后的Gabor特征进行分块,最后对分块后的图像特征再进行HOG特征提取,对提取到的HOG特征进行PCA降维,得到新的H-G特征.实验结果表明,该算法相较于其他传统单一的特征提取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.

In order to solve the problem that a single HOG feature extraction method will lose the local features of the image,a Feature extraction algorithm of Histogram of Oriented Gradients based on block Gabor is proposed.In this method,the face image to be recognized is firstly extracted by Gabor feature extraction method to get the Gabor features of the image with different scales and directions,then the Gabor features are fused according to the scale and direction,and then the fused Gabor features are introduced.Finally,the image features are extracted by HOG,and the extracted HOG features are reduced by PCA,and a new H-G feature is obtained.Experimental results show that the proposed algorithm has higher recognition accuracy and accuracy than other traditional feature extraction and recognition methods,and it is effective and robust to face interference factors such as illumination,pose expression and so on.