To solve quantification of similarity measure in the K-nearest neighbor classification, a KNN method is proposed based on LBP features and entropy-regularized Wasserstein distance, by combining the mathematical properties of Wasserstein distance in optimal mass transportation theory.Firstly, facial expression images are preprocessed.Secondly, LBP operator is applied to extract LBP feature histograms.Lastly, the K-nearest neighbor method with entropy-regularized Wasserstein distance as the similarity measure between feature histograms is used to recognize and classify facial expressions.Experimental results show that compared to the methods based on LBP only, the method greatly increases the recognition rate.%针对K最近邻分类中相似度量的量化问题,结合最优传输理论中Wasserstein距离数学特性,提出一种基于LBP特征和熵正则化Wasserstein距离的K近邻分类方法.首先对人脸表情图像进行预处理,然后使用LBP算子对图像进行特征提取获得LBP特征直方图,最后使用熵正则化的Wasserstein距离作为特征直方图之间的相似性度量的K最近邻分类方法进行人脸表情识别分类.实验结果表明该方法相较于单纯基于LBP的方法识别率有较大提高.
展开▼