Block Adjustment Method
for Mosaicing a Large Number of Satellite Image Data
2 Methodology
So as to locate GCP's, we have to identify correspondence between feature points in images and those in topographic-maps such as DCW and World Data Bank. But there are so many images where typical geographic features such as river branch points and coastal lines which employs tie-points(TP} to conjugate neighboring images and conduct geo- cording of all images simultaneously. Even if we can get enough number of GCP data, these distribution may be so biased. We propose"Block adjustment method" for mosaicing large number of data. Even if we can not find enough GCP points and these distributions are not so much homogeneous, we can adjust geometric errors over all images. The process of the Block adjustment method for mosaicing large number of data are introduced in Fig 2.
Fig.2
The Process for Mosaicing a Large Number of Satellite Image Data
We can divide this process for four stages
2.1 Measurin GCP and Tie oints for each Images
In this stage, we identify and measure coordinate values of GCP (image coordinate data ' and map coordmate data} and he pomts (Image coordmate data} for each Images. We use World Data Bank (coast and river data} to determine the map coordinate information for GCP.
2.2) Checking the Number and Distribution of GCP for each images
We use Pseudo Affine transform formula for geometric correction.
x=a1uv+a2u+a3v+a4,
y=a5uv+a6u+a7v+a8
We need more than four GCP's to determine the parameters in this formula. And GCP
points should be homogeneous. The number of GCP and these distribution have to be
checked to avoid extreme bias in GCP / TP distribution over all images.
2.3 Settin the initial value of transform araI Mters for block ad.ustIMnt and
detecting / eliminating gross errors
After checking the number and distribution of GCP points,
This stage also consists of four steps.
2.3.1} Calculate unknown transform parameters of "AA class" images (number of GCP~7}
only from GCP
2.3.2} Compute map coordinate values of TP in AA class images
2.3.3} Replace TP with GCP in images overlapping / neighboring AA class images.
2.3.4} Calculate unknown transform parameters for the neighboring images (GCP<7}
By using the information of "geo-located" tie points, we can propagate initial geo-
location of individual images from AA class image to neighboring "lower class" images.
If any gross errors may be detected through these steps, we can evaluate the original
image of GCP/TP .
2.4) Adjust geometric errors over all images with the information of tie oints
between images
By using initial values of the transfer parameters of each images, geometric errors
over all images are adjusted borrowing the idea of block adjustment in photogrametry to
compute accurate values of the transform parameters.
3. Experiment
The block adjustment method is being applied to M05-1 MESSER images. The
results will be present in the Conference.
4. Future Works
Automated identification / measurement of GCP's and TP's as to minimize human labour in GCP /TP measurement, we need to develop automated processing for identifying / measuring GCP's and TP's. For GCP identificate / measurements, feature-based matching will be employed while template matching techniques of image are applied to TP measurement.
References
-
JARS,Remote Sensing Note, Nihon Printing Co.,Tokyo,I993 pp.178 -179
- Ryosuke Shibasaki, Masataka Takagi and Koki Iwao (1995), "Land. Use Change Study of
South East Asia(LUCS-ASIA)", Proceedings of the International Symposium on
Vesitation Monitoring International Archives of Photogrammetry and Remote
Sensing, Chiba, Japan, pp.199 -205
- K.S.Rajan, Koki Iwao,Ryosuke Shibasaki & Masataka Takagi (1994), "Semi-Automated,
lassification System for Landsat Pathfinder Project", Proceedings of the 15th Asian
Conference on Remote Sensing, Bangalore, India, vol.2 pp.L-6-1 -L-6-4
