Geometric and spectral analyses of
Merged remotely sensed images
Hao-Hsiung Huang and Yi-Sheng
Hsieh
Associate Professor and Graduate Student
Department of Land Economics
National Cheng-Chi University
64, Sec. 2, Tzu-Nan Road, Wensan, Taipei, 116
Tel: (886)-2-2938-7586 Fax: (886)-2-2939-0251
E-mail: hhh@nccu.edu.tw and g7257016@grad.cc.nccu.edu.tw
CHINA TAIPEI
Keywords: Merging Images, Image Processing, Unsupervised Classification
Abstract
The research attempts to merge different types of registered remotely sensed
images by the use of simple band substitution method, color space transformation and
substitution method, and band arithmetic operation method respectively. SPOT PAN data,
SPOT XS data, and digitized aerial photography are processed for merging experiment.
Merged images were then classified using unsupervised method, and compared with each other.
The result indicates that band arithmetic operation method for merging SPOT XS data and
digitized aerial photography has the best visual display. On the other hand, color space
transformation and substitution method for merging SPOT XS data and digitized aerial
photography has the least omission error.
Introduction
Merging images obtained by different remote sensors can take advantage of the unique
characteristics of each particular data set (Chavez, 1986). Moreover, combination of image
data collected by two different systems may have information from both sensors. Therefore,
many studies(Jensen et al., 1990; Ehlers et al., 1990, Hallada, 1986; Welch and Ehlers, 1987;
Chavez and Bowell, 1988; Chavez, 1986; Grasso., 1993) have been made of digital images
merged for processing and analysis.
However, remotely sensed images acquired using different sensors usually have different
geometric and spectral resolutions. Improvement of the two resolutions may assist terrain
interpretation and mapping when merging images, thus, is the main consideration when the
method is selected for merging.
Test area and data sources, images merging, and results are described in the following
sections.
Test Area and Data Sources
Three data sets, digitized panchromatic aerial photograph, SPOT panchromatic (PAN) data, and
SPOT multispectral (XS) data shown in Figs. 1, 2, and 3 respectively, are used in this study.
The campus of National Chengchi University, located in the southeast corner of Taipei is the test
area of this study. The digitized aerial photography was ortho-rectified, and resampled to a
pixel size of 1m. The SPOT PAN and XS data sets were also ortho-rectified, and registered to
the digitized aerial photography. All data sets have the same pixel size of 1m when
co-registered to each other.
Images Merging
Three methods, simple band substitution, color space transformation and substitution, and band
arithmetic operation are tested in this study. These methods are frequently used.
Band Substitution
The SPOT PAN data is a record of both green and red energy. Therefore, it can be a good
substitute for either the green (SPOT XS1) or red (SPOT XS2) bands (Jensen, 1996). The
merged data set with SPOT XS3(near IR), SPOT PAN, and SPOT XS1 in the RGB image
memory planes, respectively, is shown in Figure 4. The merged data set with SPOT XS3,
digitized aerial photography, and SPOT XS1 in the RGB image memory planes, respectively, is
shown in Figure 5.
Color Space Transformation and Substitution
Remote sensed data presented in the RGB color coordinate system can be transformed into an
intensity-hue-saturation (IHS) color coordinate system for visual analysis. The IHS color
coordinate system can be based on a hypothetical color sphere. Intensity indicates the total
brightness of a color. Hue is the average wavelength of color. Saturation refers the purity of
the color.
The procedure shown as below (Schowengerdt, 1983) involves transforming the SPOT XS
image from the RGB to the IHS. The SPOT PAN image or digitized aerial photography is then
substituted for the intensity and the IHS transforms back into the RGB. Such an approach is
used to produce Figure 6 and Figure 7.
RGB to IHS: Three bands of remote sensor data in RGB color space are transformed into
three bands in IHS color space. The equations used in this study are as follow: (Gonzalez and
Woods, 1992)
Contrast enhancement: The image of high-spatial-resolution, (e.g., SPOT PAN data or
digitized aerial photography) is contrast stretched so that it has approximately the same
variance and mean as the intensity (I') image.
Substitution: The stretched image of high-spatial-resolution is substituted for the intensity
(I') image.
I'H'S' to R'G'B': The modified I'H'S' image is transformed back into R'G'B' color space
using an inverse IHS transformation listed as below:
Band Arithmetic Operation
It is possible to improve the geometric resolution of an image by band arithmetic operation.
The operation involves summing, differencing, multiplying, and ratioing different sources of
images. The study simply substitutes the band computed by the following equations for three
bands of the SPOT XS1 data. The results are shown in Figure 8 and Figure 9.
Results
Visual image interpretation and an unsupervised classification were employed to evaluate and
analyze the merging images.
