Collaborative representation based local discriminant projection for feature extraction

Publication Type:
Journal Article
Citation:
Digital Signal Processing: A Review Journal, 2018, 76 pp. 84 - 93
Issue Date:
2018-05-01
Metrics:
Full metadata record
Files in This Item:
Filename Description Size
1-s2.0-S1051200418300538-main.pdfPublished Version614.77 kB
Adobe PDF
© 2018 Elsevier Inc. This paper introduces a novel dimensionality reduction algorithm, called collaborative representation based local discriminant projection (CRLDP), for feature extraction. CRLDP utilizes collaborative representation relationships among samples to construct adjacency graphs. Different from most graph-based algorithms which manually construct the adjacency graphs, CRLDP is able to automatically construct the graphs and avoid manually choosing nearest neighbors. In CRLDP, two graphs (the within-class graph and the between-class graph) are constructed. Based on the two constructed graphs, the within-class scatter and the between-class scatter are computed to characterize the compactness and separability of samples, respectively. Then CRLDP seeks to find an optimal projection matrix to maximize the ratio of the between-class scatter to the within-class scatter. Experimental results on ORL, AR and CMU PIE face databases validate the superiority of CRLDP over other state-of-the-art algorithms.
Please use this identifier to cite or link to this item: