摘要： 介绍梯度方向直方图(HOG)人脸识别算法，设计基于脸部识别技术人脸库的HOG人脸识别实验，以测试不同HOG参数对人脸识别的影响，从而进行最优参数设置。实验结果表明，HOG特征在行人检测和人脸识别上对梯度方向空间和区间的选择是一致的，不同分块模式对人脸识别的影响与行人检测不同，HOG特征描述子用较少的特征维数就能有效地表达人脸，采用块内标准化方式后，识别性能有大幅度提升。

关键词: 梯度方向直方图, 人脸识别, 行人检测, 欧氏距离

Abstract: This paper describes the face recognition algorithm of Histograms of Oriented Gradients(HOG), and designs face recognition experiment based on HOG. The Experiment is done on the Face Recognition Technology(FERET) face database. It testes the effect of different HOG parameters on face recognition and tries to find the optimal parameter settings. Experimental results show that the choice of space and range in gradient direction of HOG feature is the same on pedestrian detection and face recognition. The different block mode has different effects too. HOG descriptor can express face effectively when it produce less characteristic dimension in non-overlapping manner. Recognition performance improves significantly when it is standardized.

Key words: Histograms of Oriented Gradients(HOG), face recognitionz, pedestrian detection, Euclidean distance

中图分类号: 

• TP391