Abstract
Hyperspectral imagery acquired by a hyperspectral sensor contains hundreds of narrow contiguous spectral bands, providing the opportunity to identify the various materials present on the surface. Due to the three-dimensional (3D) nature of hyperspectral data, 3D filters that could extract joint spatial-spectral features have been recently considered in the literature. In this paper, after the 3D Gabor features with certain orientations have been extracted from the raw hyperspectral image data, both the Gabor magnitude and phase features have been used for hyperspectral imagery classification, which is named as Gabor-MP. Specifically, the confidence score of each test sample is computed by support vector machine for each Gabor magnitude feature cube, while the Hamming distance is calculated based on the quadrant bit coding of each Gabor phase feature cube. Then the label of the test sample is identified by simple calculation between the confidence scores and Hamming distance values. Experimental results on two real hyperspectral data have demonstrated the effectiveness of the proposed Gabor feature fusion framework for hyperspectral imagery classification.
Bin Deng
Lecturer
My research interests include hyperspectral image processing, pattern recognition and machine learning.