Name
Abstract | ||
|---|---|---|
Geostatistical method is widely used to extract image texture for remotely sensed data classification, of which variogram function is the most important. Previous studies, however, usually calculate spatial autocorrelation in one single image and cross correlation between two images to generate texture images. They can only be applied to no more than two bands once. This paper introduces a new method to calculate image texture from multiple bands all at once. The pixels on the multichannel image are considered as vectors, while calculating image texture of multiple bands. A multivariate variogram is adopted and interpreted as a measure of distance in feature space. The experimental multivariate variogram may therefore be regarded as half the expected value of a distance, either Euclidean distance squared or spectral angle distance squared, in the multidimensional space. The multichannel texture image produced was incorporated into spectral classification. A subset of Landsat TM image was used to evaluate the proposed multichannel texture for lithological discrimination. The result indicates that compared to spectral classification, the classification accuracy can significantly be improved when multichannel texture was included in classification. The proposed methodology can be used to multitemporal and hyperspectral image analysis. © 2005 IEEE. |
Year | DOI | Venue |
|---|---|---|
2005 | 10.1109/IGARSS.2005.1525744 | IGARSS |
Keywords | Field | DocType |
classification,geostatistical texture,multivariate variogram,pixel,hyperspectral sensors,remote sensing,feature space,euclidean distance,image classification,image texture,symmetric matrices,multidimensional systems,spatial autocorrelation,cross correlation | Image generation,Computer science,Remote sensing,Artificial intelligence,Contextual image classification,Computer vision,Variogram,Pattern recognition,Multivariate statistics,Image texture,Hyperspectral imaging,Pixel,Multidimensional systems | Conference |
Volume | Issue | ISBN |
6 | null | 0-7803-9050-4 |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
2 | ||