THE NEW ORTHORECTIFICATION STRATEGY AS A WAY FOR
VISUAL QUALITY IMPROVEMENT
Krystian Pyka
Department of Photogrammetry, AGH University of Science and Technology, Krakow, Poland - krisfoto@interia.pl
KEY WORDS: Edge, Orthoimage, Orthorectification, Pixel, Quality, Visualization,
ABSTRACT:
In the paper a new orthorectification strategy of aerial images is presented. The strategy is focused on improvement of visual quality
of orthoimage. The idea of proposed strategy relies on extraction of edges from the image and next special resampling is taken. The
critical issue for edge detection is that the images usually includes some noise. The de-noising of an image has many solutions but
none of them is perfect. Therefore in the paper the usefulness of wavelet transformation for image de-noising is tested.
The proposed orthorectification strategy gives a better image then the image from typical orthorectification. The visual quality of
edges in orthoimage is comparable to the original image.
1. INTRODUCTION
The paper includes a proposal of a new orthorectification
strategy of images.
The geometric transformation of images especially the
transformation between the central perspective image and the
orthoimage is performed in the same way for many years. The
only changes that have been implemented, rely on the
resignation from any simplification in the computing process.
Now more often the exact transformation using collinearity is
computed for all pixels and breaklines of the terrain are taken
into account. These changes are sufficient to achieve the good
geometric quality, but the visual quality of orthorectified image
is always worse than the visual quality of the original image.
The reason why visual quality is reduced is that
orthorectification is the same both for linear elements and for
homogenous areas. In the typical rectification the intensity
value of an orthoimage pixel is a result of interpolation of the
intensity values of a group of image-pixels (from one to
sixteen). Therefore on the intensity of pixel on the edges of a
road is influenced by the intensity of a road surface and
shoulder.
The proposed modification of orthorectification strategy is
focused on the improvement of visual quality of an orthoimage.
The main steps of the proposed strategy are: (1) image de-
noising using wavelet transformation, (2) detection of edge
using Laplace operator, (3) the execution of orthorectification
with a special resampling method.
2. IMAGE DE-NOISING USING WAVELET
TRANSFORMATION
Aerial and satellite images usually contain noise, due to image
acquisition, transmission errors or compression side effects. The
noise causes great problems to image processing algorithms.
(Pitas, 2000). Only when the image de-noising is effective the
edge detection is proper. In other case the edge detection is
significantly noised.
De-noising is typically handled by smoothing filters. But
smoothing can delete useful information or distort the input
image. The mainly used smoothing method is the Gaussian
2.D filter. The Gaussian outputs are an average of each pixel's
neighborhood weighted more towards the value of the central
pixels. In comparison with a mean filter, the Gaussian provides
a gentler smoothing and preserves the edges better. But the
Gaussian filter works well only in isotropic spaces therefore
when it is used for smoothing real images it removes both noise
and detailed information.
An other solution to de-noise is the usage of the wavelet
transformation. This transformation separates the smooth
variations and details of the image much better than Fourier-
based techniques.
Wavelet transformation is widely applied to image compression
and is more often used for image de-noising (Rangarajan,
2002).
Wavelet transformation decomposes a discrete image into four
sub-image where LL, HL, LH, HH represent average, vertical
fluctuation, horizontal fluctuation and diagonal fluctuation
respectively. The decomposition may be held on several levels
(Walker, 1999). A sample of all the sub-images on the first
level is shown in the figure 1b. As wavelet transformation the
Haar function was used.
The basic method of image de-noising is thresholding. By
choosing a correct threshold, it is possible to remove most of
the random noise. In this research a soft thresholding adequate
for each sub-image was tested:
(76) FG ZT
fe. DA ifGj) 2T2 AND (la) (1)
| 0 GN «TO
the extra condition (1a) are as follow:
^
for LH: if 1767 - HIST] OR |/f(i,j40| 27l
for HL: if |f(i-1,5)| =T! OR | fU+L Di 271
for HH: f£ lj/ü-Lj-D| 271 OR I/6-1;«D[zml
OR |7(+1,j+0| 271 OR |fu+lj-D|20
Inte
whe
The
sub
dec
con
thre
dire
det:
a €
thre
Ed
cor
Th
of
ma
int
20
Ag