Page 145 - Digital Analysis of Remotely Sensed Imagery

114 Cha pte r F o u r

rectify it using polynomial, rubber sheeting, or a predefined model
for IKONOS and SPOT imagery. Such rectified images may be
mosaicked. In generating a controlled mosaic from overlapping
aerial photographs, options (e.g., minimum, maximum, average)
are available to specify how the output image should look like
radiometrically over the overlapped portion. It is also possible to
import/export cut lines, and smooth images along cut lines, and to
balance imagery’s color using dodging in Imagine.

4.2.3 Image Enhancement
All kinds of image enhancement, be it radiometric, spatial, or spec-
tral, are performed under Imagine Interpreter. Radiometric correc-
tion refers to the removal of atmospheric effects (e.g., haze) from the
input image so that its pixel values correspond closely to the reflec-
tance of targets on the Earth’s surface. Radiometric enhancement
includes contrast enhancement, haze and noise removal, inversion of
brightness, and destripping (for TM imagery only). Contrast of an
image may be enhanced through a look-up table and histogram-
based manipulation (e.g., histogram equalization and histogram
matching). Image enhancement in the spatial domain includes tex-
ture analysis, focal analysis, and image convolution. Spectral enhance-
ments may be carried out via principal component analysis, Fourier
analysis, image transformation from hue-intensity-saturation to red-
green-blue or vice versa, and image indexing. With version 9.1, it is
possible to pan-sharpen low resolution multispectral bands with a
finer resolution panchromatic band. The quality of the sharpened
image may be improved via a two-pass filtering option.
Specific tools (e.g., band normalization, spectrum averaging, pro-
filing, spectral library, and so on) have been developed to handle
hyperspectral data. Also found in the Interpreter module are two ana-
lytical functions for both remote sensing and non-remote sensing data:
GIS analysis and topographic analysis. Some GIS functions (e.g.,
clumping and sieving) are essential in performing postclassification
processing such as spatial filtering and thematic generalization.

4.2.4 Image Classification
ERDAS Imagine supports both unsupervised and supervised classifi-
cation. The unsupervised classification algorithm is called Iterative
Self-organizing Data Analysis Technique (ISODATA). Images may be
classified using one of the four supervised methods: parallelepiped,
minimum distance, maximum likelihood, and mahalanobis. Special
tools are available for selection of training samples and for analyzing
their separability. Unique to ERDAS Imagine is its Knowledge Classi-
fier that allows multiple decision rules to be combined logically to
deduce the likely identity of a pixel in question. These rules are con-
tained in a knowledge base that is created via the Knowledge Engine.
All classified results may be assessed for their accuracy using the