基于LTS优化图像切边线性回归分类的人脸识别研究

针对现实人脸识别中由于伪装(如围巾、太阳镜和头发)或其它物体引起的面部遮挡而严重影响识别率的问题,提出了一种基于最小截平方和的图像切边线性回归分类算法。首先,使用一个鲁棒性强的估计量检测并裁剪查询样本、训练样本中受污染的像素点；然后,利用线性回归分类算法对图像进行切边；最后,利用LTS计算出规范化的重构误差。实验结果表明,相比其它几种回归分类算法,本文算法取得了更高的识别率,同时大大降低了训练总完成时间。

Research of Image Trimming LRC Optimized by LTS in Face Recognition

There are usually facial occlusion generated by masks such as scarf in truly face recognition, for the issue that the mask will impact recognition accuracy, linear regression classification algorithm with image trimmed based on least trimmed squares is proposed. Firstly, a robust estimator is used to cut polluted pixels of query and training samples out. Then, linear regression classification algorithm is used to trim images. Finally, reconstruction error is compute. Experimental results show that proposed algorithm has higher recognition accuracy and less training time than several advanced regression classification algorithms.