基于中心域的LPP算法研究

Research on locality preserving projections based on center field

针对LPP算法中最近相邻图不能很好地表示流形的局部结构问题,提出一种基于中心域的保局投影算法。该算法采用LBP获取图像的高阶全局统计信息,并将其投影到LPP的流形空间。流形空间的获取是以各图像间的中心域的欧式距离为标准构建最近相邻图,使其可以简单地、较好地表示流形局部结构,并得到数量较少的特征维数。在ORL人脸库上的实验结果表明,该方法可有效地降低特征维数,并取得较好的识别率。

In order to solve the problem that the nearest neighbor graph can not always accurately estimate the local manifold structure,a new algorithm of locality preserving projections based on the center field is presented.The algorithm used LBP to obtain the high-order statistical information,and then projected onto the manifold LPP space.To get the manifold spaces is to construction of nearest neighbor graph based on the Euclidean distance of center field between image,So that it can be better and stable express the local manifold structure,and get a smaller number of feature dimensions.The results of the comparative experiment on the ORL face images indicate that the proposed method can effectively reduce dimension and achieve a higher recognition rate.