Abstract: The existing face recognition algorithms lack the ability of automatic regulation for illumination variation. This paper describes a face recognition algorithm based on both global and local feature. Principal Component Analysis is performed to extract global features. A special strategy is used to combine different local features according to their image entropy. Bayes integration is adopted to fuse both global and local features and the final result is given. The experiments demonstrate that this algorithm can combine global and local information well and improve the efficiency of the face recognition rate.

Key words: face recognition, local feature, global feature, Principal Component Analysis（PCA）, image entropy, Bayes theory

摘要： 针对人脸识别算法缺乏对光照变化的自动调节能力的弱点，提出了一种综合利用全局和局部特征进行人脸识别的新方法。对整幅人脸图像进行主成分分析；针对人脸局部特征，提出一种根据各局部子块包含的信息量即利用图像熵值进行自动加权的算法；基于贝叶斯原理对全局和局部特征进行数据融合，给出最终结果。实验结果表明，该数据融合算法综合全局和局部特征提取方式的优势，有效提高了人脸识别率。

关键词: 人脸识别, 局部特征, 全局特征, 主成分分析, 图像熵, 贝叶斯原理