Geography Reference
In-Depth Information
These tie points tie the line/pixel positions in image coordinates to geographical
latitude/longitude and can be used as ground control points (GCP) to register an
image to a geocoded target image. This work first created a master geocoded image
with the same resolution as the RADARSAT-2 images and then registered the
RADARSAT-2 images to this geocoded master image using PCI Geomatica based
on the tie points. Visual inspection indicates that the RADARSAT-2 images were
registered perfectly.
19.3.2
Object-Oriented Image Analysis of RADARSAT-2
PolSAR Images
Most change detection and classification methods for remote sensing images are
pixel-based. When applied to PolSAR images, pixel-based methods have two
disadvantages. First, they are prone to be affected by speckles in PolSAR images
and produce errors (Qi et al.
2012
). Second, they are difficult to use to extract and
utilize spatial and textural information, which is helpful in improving classification
accuracy of remote sensing data (Gao et al.
2006
). Object-oriented image analysis
can be used to reduce the speckle effect and extract textural and spatial informa-
tion to support PolSAR image classification. In object-oriented image analysis,
image objects (groups of pixels) are first delineated by using image segmentation
techniques, and then change detection and classification are implemented on the
object level. In comparison with pixels, image objects are little affected by speckles.
Moreover, object-oriented image analysis enables the acquisition of a variety of
textural and spatial features for improving classification accuracy. A feature is an
attribute that represents certain information concerning objects of interest. Given
that regions in an image provide considerably more information than do pixels,
many different features for measuring color, shape, and texture of the associated
regions can be used (Benz et al.
2004
).
In this study, the object-oriented package Definiens Developer 7.0 (Baatz et al.
2004
) was used to implement the object-oriented image analysis of RADARSAT-2
images. The multi-resolution segmentation module provided by Definiens Devel-
oper 7.0 was used to perform image segmentation based on shape and color
homogeneity. For change detection, a straightforward approach for image seg-
mentation is to segment images acquired at different times separately and then
overlay them together. However, this method will produce inconsistencies in the
delineation of boundaries of objects and result in a large number of fragmental
patches in the final segmented image (Li et al.
2009
). Such excessive fragmentation
can lead to difficulties in change detection. A hierarchical image segmentation
procedure was proposed to minimize the inconsistency in delineating objects from
two successive images. In considering two co-registered images, image (
t
1
)and
image (
t
2
), acquired over the same area at different times
t
1
and
t
2
, the procedure of
hierarchical segmentation can be summarized as follows: (1) the initial segmentation
is applied to image (
t
1
) with a fixed scale parameter; and (2) the same segmentation
Search WWH ::
Custom Search