仿射变换下的分类识别

特征提取和分类识别是统计模式识别中两大关键步骤。显然,不同的特征提取方法与不同的分类器相结合,识别性能往往是不同的。从微分几何的角度出发,可将特征系数的获得看成线性几何变换,即仿射变换,据此在黎曼空间提出一种基于黎曼度量的分类识别方法。通过对经典最近邻分类器的线性加权,达到更有效地分类识别。不但在理论上将特征系数提取与分类识别合理的结合起来,而且由人脸识别实验表明该方法的有效性,该方法比传统方法的识别率有约 3%的提高。

Classification and recognition through affine transformation

Feature extraction and classification recognition are two critical steps in pattern recognition. The results obtained by different methods of feature extraction combining with different classifiers are not the same. From the view of differential geometry, the extraction of feature coefficient can be regarded as a linear geometrical transformation, i.e. affine transformation. Thus a classifier based on Riemannian metric is proposed. By weighting the nearest neighbor classifier linearly, this method can do better classification and recognition. The method not only reasonably combines the feature coefficient extraction with classification recognition theoretically, but also presents effectiveness demonstrated by face recognition experiments. The mean recognition rate of the proposed method is about 3% higher than that of the traditional method, which shows that the presented method is effective.