Research Methodology - Historical Land Use/Land Cover Classification Using Remote Sensing - page 66

periods of times, or they were obtained from two or more different sites; and
rectifying an image to be accurate to an individual coordinate system (geo-coding)
(Liu and Mason 2009 ). ''Spatial distortion arises from scanner characteristics and
their interaction with the airborne platform or satellite orbital geometry and figure
of the Earth'' (Schowengerdt 2007 ). Geometric correction can maximize the
usefulness of the remotely sensed data for information extraction (e.g., thematic
maps).
The geometric correction is the first image processing step (pre-classification
approach) carried out when the remotely sensed data are not geo-rectified (Liu and
Mason 2009 ). However, geo-rectification can be carried out as a post-classification
approach to reduce the errors and distortions resulting from the geometric cor-
rection process. Generally, it is more competent to begin with geo-rectifying the
still unprocessed data. Therefore, all products that will result from the raw data
will be automatically geo-rectified (Liu and Mason 2009 ).
The problems that can occur in pixels of an image that will be rectified to other
one (source image) are: the pixels have a different position; different orientation;
and different size (Fig. 5.2 ). With this in mind, resampling methods have been
developed to cope with these problems. The methods are based on choosing well-
known and matching sites in both images of the selected cartographic projection.
Based on these sites, a resampling technique will calculate the relation between
their positions in the two images. These positions can be located exactly on an
image using the so-called Ground Control Points (GCPs). These points are
potential to define a suitable transfer function to be applied between the both
images, i.e. rectify and master scenes (McCloy 1995 ). There are three components
to the process: (1) selection of suitable mathematical distortion model(s); (2)
coordinate transformation; and (3) resampling (interpolation). These are also
known as warping (Wolberg 1990 ).
'' Resampling is the process of calculating the data file values for the pixels in
the rectified image by the use of data file values in the source image data''
(McCloy 1995 ). There are three resampling schemes: nearest neighbor (sometimes
called zero-order interpolation); bilinear interpolation; and cubic convolution. In
the nearest neighbor approach, ''the data file value of the nearest pixel to the
retransformed pixel in the source image is adopted as the data file value for the
output rectified pixel'' (Liu and Mason 2009 ). By comparison with the other two
schemes, it has the advantages: that it does not change the digital number value in
Fig. 5.2 Reposition pixels
from their original locations
(input matrix) in the data
array into a specified
reference grid (output matrix)
(Source modified from
Lillesand et al. 2008 )