基于最大边界准则的稀疏局部嵌入特征提取方法

发布时间：2018-05-21 20:44

本文选题：特征提取 + 局部线性嵌入　；参考：《计算机应用研究》2017年05期

【摘要】：针对局部线性嵌入算法(LLE)无法利用样本的鉴别信息,最大边界准则算法(MMC)对非线性数据处理能力较弱的不足,提出一种基于最大间距准则的稀疏局部嵌入(SLE/MMC)特征提取算法。在保持局部近邻的前提下,首先在类内紧致图中使同类样本尽可能地聚集在一起;然后在类间惩罚图中使不同类别的样本尽可能分离;最后使用弹性网回归算法得到一个最优的稀疏投影矩阵。为了避免小样本问题,采用MMC的形式构造目标函数。在ORL、Yale以及UMIST人脸库上的实验结果可以看到,同其他几种方法(PCA、LLE和MMC)相比,SLE/MMC具有更高的识别率,表明该方法具有更高效的特征提取能力。
[Abstract]:Because the local linear embedding algorithm (LLE) can not utilize the discriminant information of the sample and the maximum boundary criterion (MMC) algorithm has a weak ability to deal with nonlinear data, a sparse local embedding / MMC feature extraction algorithm based on the maximum spacing criterion is proposed. On the premise of preserving local nearest neighbors, the same kind of samples are gathered together in the compact graph of class as much as possible, and then the samples of different classes are separated as far as possible in the punishment graph between classes. Finally, an optimal sparse projection matrix is obtained by using elastic network regression algorithm. In order to avoid the small sample problem, the objective function is constructed in the form of MMC. The experimental results on ORL Yale and UMIST face database show that the proposed method has a higher recognition rate than other methods such as PCALE and MMC, which indicates that the proposed method is more efficient in feature extraction.
【作者单位】：南昌航空大学江西省图像处理与模式识别重点实验室;南昌航空大学信息工程学院;南京理工大学高维信息智能感知与系统教育部重点实验室;南京晓庄学院可信云计算与大数据分析重点实验室;
【基金】：国家自然科学基金资助项目(61462064,61272077,61203243,61262019,61362031) 高维信息智能感知与系统教育部重点实验室(南京理工大学)基金资助项目(30920140122006) 中国博士后基金资助项目(2014T70453,2013M530223)
【分类号】：TP391.41

【相似文献】