Spatial segmentation of multi/hyperspectral imagery by fusion of spectral-gradient-textural attributes

Abstract

We propose an unsupervised algorithm that utilizes information derived from spectral, gradient, and textural attributes for spatially segmenting multi/hyperspectral remotely sensed imagery. Our methodology commences by determining the magnitude of spectral intensity variations across the input scene, using a multiband gradient detection scheme optimized for handling remotely sensed image data. The resultant gradient map is employed in a dynamic region growth process that is initiated in pixel locations with small gradient magnitudes and is concluded at sites with large gradient magnitudes, yielding a map comprised of an initial set of regions. This region map is combined with several co-occurrence matrix-derived textural descriptors along with intensity and gradient features in a multivariate analysis-based region merging procedure that fuses the regions with similar characteristics to yield the final segmentation output. Our approach was tested on several multi/hyperspectral datasets, and the results show a favorable performance in comparison with state-of-the-art techniques.

Discriminating Spectral–Spatial Feature Extraction for Hyperspectral Image Classification: A Review

“…Spectral features, which are generally extracted from each spectra/pixels of HSI cubes, can be common features, such as spectral amplitude [17][18][19], spectral gradient [20][21][22][23], global tendency [36], etc. However, these manual features are generated by the feature descriptors designed empirically, which is not always effective and robust for complex HSI data sets.…”

“…Traditional Feature Descriptors Spectral features are the implicit peculiarities and patterns of each spectrum of HSI and the foundation of classification. Shallow spectral features generally obtained by traditional feature descriptors include spectral amplitude [17][18][19], spectral gradient [20][21][22][23], global tendency [36,38], local variance [36][37][38][39], etc. Spectral amplitude is the original grayscale value of a spectrum.…”

“…In the early days, traditional methods mainly focused on the extraction of hand-crafted features. For example, spectral amplitude [17][18][19] and gradient [20][21][22][23], as intuitive metrics, were exploited to characterize the spectral intensity and its variation. Histograms of oriented gradient (HOGs) [24][25][26][27] and local binary patterns (LBPs) [28][29][30] were effective tools to capture local shapes and textures.…”

Evaluation of gradient operators for hyperspectral image processing

“…The basic strategies approximate multivariate gradient through a marginal approach, followed by combining the marginal results with a maximum function in (4) or weighted sum [16]. The problem with a marginal approach lies in the fact that it does not preserve metrological constraints and is less sensitive to gradient coherence between the spectral bands [7].…”

Texture Based Superpixel Segmentation Algorithm for Hyperspectral Image Classification

“…Segmentation is another popular approach for inclusion of spatial features [13]. With the traditional fixed window-based methods, the occurrence of salt and pepper noise in the classification result is quite prevalent.…”