Hyperspectral classification using Markov random field and “spatial probability density function”

Hyperspectral images have the unique ability to provide both a spatial sampling and a spectral sampling. Although the hyperspectral data contain a lot of information about the spectral properties of the land cover, but no spatial information is inherent in the spectral data. This problem can be solved using a joint spectral/spatial classifier. In this paper we propose a classification algorithm based on Markov random field and using spatial and spectral information simultaneously. In proposed algorithm spectral features are extracted at first step. At second step an iterative spatial-spectral classification algorithm is applied. ldquospatial probability density functionrdquo of classes is estimated using kernel density estimation in second step. The hyperspectral data set used in our experiments is a scene taken over NW Indianapsilas Indian Pine by the AVIRIS sensor. The obtained results show that proposed classifier improved classification accuracy significantly.