Table of Contents Author Guidelines Submit a Manuscript
Mathematical Problems in Engineering
Volume 2013 (2013), Article ID 367105, 10 pages
Research Article

The Study of Scene Classification in the Multisensor Remote Sensing Image Fusion

Ji Li1,2 and Zhen Liu1,2

1College of Computer Science, Chongqing University, 400030 Shapingba, Chongqing, China
2Key Laboratory for Dependable Service Computing in Cyber Physics Society of Ministry of Education, China

Received 24 March 2013; Accepted 28 April 2013

Academic Editor: Hua Li

Copyright © 2013 Ji Li and Zhen Liu. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.


We propose a scene classification method for speeding up the multisensor remote sensing image fusion by using the singular value decomposition of quaternion matrix and the kernel principal component analysis (KPCA) to extract features. At first, images are segmented to patches by a regular grid, and for each patch, we extract color features by using quaternion singular value decomposition (QSVD) method, and the grey features are extracted by Gabor filter and then by using orientation histogram to describe the grey information. After that, we combine the color features and the orientation histogram together with the same weight to obtain the descriptor for each patch. All the patch descriptors are clustered to get visual words for each category. Then we apply KPCA to the visual words to get the subspaces of the category. The descriptors of a test image then are projected to the subspaces of all categories to get the projection length to all categories for the test image. Finally, support vector machine (SVM) with linear kernel function is used to get the scene classification performance. We experiment with three classification situations on OT8 dataset and compare our method with the typical scene classification method, probabilistic latent semantic analysis (pLSA), and the results confirm the feasibility of our method.