Abstract: In this paper，a new face recognition technique based on block universal ICA（BUICA） is proposed.First，the original images are divided into blocks in proposed method，then ICA could be directly used to all the blocks to construct the eigenspace，at last the blocks of one images are projected to the eigenspace to feature extract and recognition.BUICA could reduce the image dimension and extract local features effectively.The experimental results on YALE and FERET face database indicate that the performance of BUICA is superior of that ICA.

Key words: ICA, BUICA, feature extraction, face recognition

摘要： 提出分块独立成分分析的特征抽取方法，并成功应用于人脸识别。分块独立成分分析方法先对图像矩阵进行分块；然后对所有图像子块联合进行独立成分分析，构造特征空间；最后把图像所有的子块投影到特征空间提取特征进行分类识别。其特点是可以有效降低图像维数和有效提取图像局部特征。在YALE和FERET人脸库上的实验结果表明，提出的分块独立成分分析方法明显优于独立成分分析方法。

关键词: 独立成分分析, 分块独立成分分析, 特征抽取, 人脸识别