The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
356
also shadows of buildings around these roads lie on them. In a
low resolution image the size of pixel could be bigger than the
width of road so the lower the image resolution the harder to
detect the roads (Figure 2).
Figure 2: Roads that extracted by using automated methods.
2.2. Semi- Automated Methods
Filtering
Spatial frequency is described as the number of variations
among pixels’ values in a specific region over the raster dataset.
If this variation is low that image can be described as low
frequency image, otherwise if variation is high the image
described as high frequency image.
High pass filters are used for increasing the spatial frequency of
images while Low pass filters are used for reducing or
suppressing the spatial frequency of images. In filtering
processes each pixel evaluated with particular number of its
neighbors pixels and depends on weight the new value of each
pixel is computed and then assigned to each one.
PREWITT FILTER
SOBEL FILTER
Quickbird
Quickbird
Ikonos
Ikonos
Aster
Figure 3: High pass filtered (Prewitt and Sobel) images.
Classification
In this study in order to detect the roads edge detection filters
which are the types of high pass filters (Prewitt and Sobel) are
conducted to each satellite images having different spatial
resolution. In this way spatial frequency of each image was
increased so the roads were highlighted and then extracted from
each image (Figure 3). According to this processes’ results, in
high resolution images such as Quickbird and Ikonos the main
and secondary roads are detected more accurately, however in
low resolution images (Spot, Aster and Landsat-ETM) the
details are reduced and detection of roads become more difficult
therefore at some region only the main roads could be detected.
Classification can be described as, grouping image pixels into
categories or classes to produce a thematic representation.
Classification can be used in thematic maps or can be further
incorporated into digital analysis. It can be performed on single
or multiple image channels to separate areas according to their
different scattering or spectral characteristics. Digital image
classification procedures are differentiated as being either
supervised or unsupervised.
In this study Maximum Likelihood classification technique was
applied four images (Quickbird, Ikonos, Aster, and Landsat)