کد مقاله | کد نشریه | سال انتشار | مقاله انگلیسی | نسخه تمام متن |
---|---|---|---|---|
411618 | 679578 | 2016 | 8 صفحه PDF | دانلود رایگان |
• We points out the shortcoming of NPE in learning the local structure.
• Our approach integrate both global and local structure of data.
• Characterizes the Euclidean distance and geometric reconstruction relationship of data.
Neighborhood preserving embedding (NPE) has been widely used to learn the intrinsic structure of data. However, it may impair the local topology and ignore the diversity of data. In this paper, we present a dimensionality reduction approach, namely discriminant neighborhood structure embedding (DNSE). DNSE constructs an adjacency graph to characterize the diversity of data and combines NPE to learn the local intrinsic geometric structure, which well characterizes both similarity and diversity. Finally, the global structure, which is obtained by LDA, is integrated with the aforementioned local structure to build the objective function. Experiments on the four image databases illustrate the effectiveness of the proposed approach.
Journal: Neurocomputing - Volume 174, Part B, 22 January 2016, Pages 850–857