【摘要】 提出一种非监督K均值聚类的人脸识别方法,该方法首先根据人脸结构特征将人脸图像矩阵分块,接着自适应地计算每个分块在分类中的权值,最后根据类别的权值大小进行分类。实验中采用了Orl和Yale人脸库和最近邻分类器测试该方法,测试结果表明,此方法有效,且对光照和人脸表情具有很好的鲁棒性,较传统的经典人脸识别法具有更高的识别率。更多还原

【Abstract】 The weighted modular facial recognition approach based on K mean clustering method(WMKC) is presented in this paper,which firstly partitions an original whole image into subpatterns according to facial features,then adaptively computes the weight of each subpattern in classification,and finally classifies subpatterns in terms of their weight in each classification.In this experiment,the Yale and the Orl facial image database are used to test this algorithm by using the nearest neighborhood algorithm to construct the classifiers.Experimental results show that the proposed method is effective,robust to both facial expressions and illumination variations,and superior to the traditional facial recognition approach in recognition rate.更多还原