摘要
提出了基于Grassmann流形的半监督图像集鉴别分析方法。该方法将子空间表示成Grassmann流形上的点,分别用一组单位正交基表示。通过Grassmann核函数,度量子空间的相似度。不同于其他基于Grassmann流形的图像集鉴别分析,引入图嵌入框架,通过保持数据局部邻域结构的同时,最大化不同类别数据的距离,得到最优投影矩阵,并在投影空间中进行图像集分类。采用半监督学习,对于未标记样本,根据其最近邻类别进行估计。实验表明,该方法取得了优于其他图像集识别算法的效果。
A semi-supervised discriminant analysis based on Grassmann manifold is presented for image set matching. Each image set can be viewed as a point on the Grassmann manifolds represented by an orthonormal matrice. Through Grassmann kernel, the similarity of two subspaces could be evaluated and discriminant analysis could be easily extended to images sets. Different from other methods on Grassmann manifold, the local structure of data is taking into account. Through comparison experiments, it is shown to be generally better than other image sets matching methods.
出处
《微型电脑应用》
2012年第6期19-22,共4页
Microcomputer Applications
基金
国家自然科学基金NSFCNo.60775009
