Hyperspectral Image Classification by Nonlocal Joint Collaborative Representation With a Locally Adaptive Dictionary

• Published in: IEEE transactions on geoscience and remote sensing Vol. 52; no. 6; pp. 3707 - 3719
• Main Authors: Li, Jiayi, Zhang, Hongyan, Huang, Yuancheng, Zhang, Liangpei
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.06.2014; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Classification; Collaboration; Construction; Consumption; Dictionaries; hyperspectral imagery; Hyperspectral imaging; Image classification; joint collaboration model; Joints; k -nearest neighbor (K-NN); Pixels; Representations; sparse representation; Sparsity; Support vector machines; Training; Vectors; k-nearest neighbor (K-NN); joint collaboration model; Classification; hyperspectral imagery; sparse representation
• ISSN: 0196-2892; 1558-0644
• DOI: 10.1109/TGRS.2013.2274875

Sparse representation has been widely used in image classification. Sparsity-based algorithms are, however, known to be time consuming. Meanwhile, recent work has shown that it is the collaborative representation (CR) rather than the sparsity constraint that determines the performance of the algorithm. We therefore propose a nonlocal joint CR classification method with a locally adaptive dictionary (NJCRC-LAD) for hyperspectral image (HSI) classification. This paper focuses on the working mechanism of CR and builds the joint collaboration model (JCM). The joint-signal matrix is constructed with the nonlocal pixels of the test pixel. A subdictionary is utilized, which is adaptive to the nonlocal signal matrix instead of the entire dictionary. The proposed NJCRC-LAD method is tested on three HSIs, and the experimental results suggest that the proposed algorithm outperforms the corresponding sparsity-based algorithms and the classical support vector machine hyperspectral classifier.