Classification of High Resolution Images with Different Cues by using CRF Model

Abstract

In High-resolution images sematic labeling perform
very important task for classification. Sematic labeling is done
with the set of pixels. And it is also known as Semantic
Segmentation. Here the predictable algorithm integrates spectral,
spatial contextual and spatial location cues by modeling the
probabilistic potentials.Two common words are used in cues while
integrating different potentials i.e. Spatial and Spectral. Spatial
extracts objective which is set of pixels. And Spectral extracts
additional information which human eye fails to capture.
Experiments Hyper-spectral images display strong dependencies
across spatial and spectral neighbours, which have been
proved to be very helpful for hyper-spectral image classication.
High- resolution images have the type of plentiful geometric and
detail information, which are useful to detailed classication A
classication algorithm based on conditional random elds (CRFs)
is presented algorithm integrates spectral, spatial contextual and
spatial location cues by modeling the probabilistic potentials.
The spectral cues modelled by the unary potentials can give
basic information for perceptive the various land-cover classes.
The pair-wise potentials believe the spatial contextual information
by establishing the neighbouring connections between pixels to
favour spatial smoothing. The spatial location cues are explicitly
set in the higher order potentials. The higher order potentials
consider the nonlocal range of the spatial location interactions
between the target pixel and its nearest training samples. This
can give useful information for the classes that are easily confused
with other land-cover types in the spectral appearance. The
CRFSS algorithm integrates spectral, spatial contextual and
spatial location cues within a CRF framework to provide opposite
information from altering perspectives, so that it can address the
common problem of spectral variability in remote sensing images,
which is directly reected in the accuracy of each class and the
average accuracy.