HSI are 3D data, with a spectral signature for the scene spread over several bands. Traditionally, the high dimensional spectral information is used to perform a pixel-by-pixel classification of the scene. Band subset selection/feature extraction methods have been developed to improve the performance of parametric classifiers. However, the classification accuracies of these methods are not satisfactory. In this poster, classification of HSI using spatial features and a distance classifier is presented. The features capture texture characteristics over spatial extents from the images. Texture cues are widely used for gray scale/color image processing and play a significant role in biological visual processing as well as in machine based computer vision. Texture measures quantify image characteristics such as contrast, size of elements, their distribution (homogeneous, regular or random), directionality (orientation), and symmetry. From 3-dimensional data, texture can be used to determine shape from shading, or ntation and other structural details that leads to object recognition/identification. Results show that spatial texture features results in higher classification accuracies for vegetation, urban, natural, man made objects and benthic classes compared to using only spectral features. Edge based texture and multiresolution features have been used. The classifier metric also plays an important role in improving performance. Recently, Support Vector Machines (SVM) have been shown to perform better for HSI classification using the complete spectral signatures. Hence, results are also presented using SVM instead of a distance classifier with wavelet features.


Poster presented at the 2006 Thrust R2C Multi Spectral Discrimination Methods Conference


Texture Contextual Information, Extraction, HSI

Subject Categories

Three-dimensional imaging


Biomedical Engineering and Bioengineering


Bernard M. Gordon Center for Subsurface Sensing and Imaging Systems (Gordon-CenSSIS)

Publication Date


Rights Holder

Bernard M. Gordon Center for Subsurface Sensing and Imaging Systems (Gordon-CenSSIS)

Click button above to open, or right-click to save.