Spatial-Spectral Feature Extraction of Hyperspectral Images using Attribute Profile With Partial Reconstruction and 3-D Gabor Filter Bank

Simultaneously spatial-spectral feature extraction is preferred for classification of remotely sensed images. In this paper, attribute filters with partial reconstruction are applied for extraction of spatial characteristics of hyperspectral images. In addition, the 3-D Gabor filter bank is used for simultaneously extraction of spatial-spectral features. For implementation of attribute profiles with partial reconstruction, the first 3 principal components of data and 3 types of attributes (area, inertia, and standard deviation) are used for spatial features extraction. For extraction of 3-D Gabor features, firstly the 3-D Gabor filter bank is applied to the data. Then, the spatial and spectral properties in the first 20 principal components output are computed for classification. In the proposed method, the principal component analysis transform is used for dimensionality reduction and avoiding the complexity of the calculations. The support vector machine is used as classifier where its input is the spectral-spatial features extracted by attribute filters with partial reconstruction and 3-D Gabor filters.