3.3.2 Methods Used for Documenting City Growth
At the beginning of the analysis digital editing of aerial and satellite images
(Geomatica, PhotoShop software) and image data transformation using the Topol
software was conducted. Automatic classification may not be suitable for black and
white aerial or satellite images, and this is especially true when a poor quality image
is employed. Based on the extension of textural signatures the classification has
been improved, but this technology seems not to be able to provide a fine classification. A primary visual interpretation using the Topol software was done. It runs in
two steps – vectorisation and interpretation. Homogenous areas were manually
vectorised. For the interpretation a simplified and modified classification key from
the Corine Land Cover was elected. In this case it is primarily a demonstration of
research possibilities of urban growth (Muhmmad 2004).
The method of single-image photogrammetry with collinear transformation was
used (Pavelka 2001). When more than four control points were found and
implemented, the adjustment at individual point deviations from the ideal state
was found. Achieved deviations varied between 15 and 20 m on the ground control
points for the declassified satellite images from a relatively flat area (theoretical
geometrical resolution depends on the quality and satellite system; it can be 10 m or
better for high quality and resolution images). Better results cannot be achieved
without image orthogonalization. Due to the character of this project the absolute
accuracy is not the main factor. The main objective was to evaluate the content of
images on a case project that demonstrates the potential of this technology.
The main challenge was the image quality, which varies greatly. Images from
the 1970s have been worse from a radiometric point of view even after digital
image processing. In terms of evaluating the condition and type of vegetation
unambiguously these images are not suitable for such work because panchromatic range does not give a good alternative to separate classes using classification and the number of classes is very hard to recognize. Either additional
information about the landscape (e.g. archive maps) or multispectral images
should be used. Unfortunately these images are not available for older data.
The classification of high quality black and white aerial photographs into core
classes (various types of built-up areas, roads, fields, low vegetation outside the
agricultural fields, forests of different ages, isolated trees, water) achieved an
accuracy of 86–93 % class depended. Experiments were made using recognition
software based on object-oriented classification with help of newly calculated
channels (Halounova ´ 2004a, b, c).
Theoretically, it would be logical to start from the oldest images and to expand
the vector database to rectify based on visible changes. The procedure was necessary to reverse because the latest images are of the best quality. There is a sufficient
amount of additional material and the possibility of multispectral images for the
new images. It is also possible to assume that the character of some essential parts
hasn’t changed fundamentally. The final task was to choose a suitable interpretive
key from the Corine Land Cover.
48
K. Pavelka and E. Matous ˇkova ´
At the beginning of the analysis digital editing of aerial and satellite images
(Geomatica, PhotoShop software) and image data transformation using the Topol
software was conducted. Automatic classification may not be suitable for black and
white aerial or satellite images, and this is especially true when a poor quality image
is employed. Based on the extension of textural signatures the classification has
been improved, but this technology seems not to be able to provide a fine classification. A primary visual interpretation using the Topol software was done. It runs in
two steps – vectorisation and interpretation. Homogenous areas were manually
vectorised. For the interpretation a simplified and modified classification key from
the Corine Land Cover was elected. In this case it is primarily a demonstration of
research possibilities of urban growth (Muhmmad 2004).
The method of single-image photogrammetry with collinear transformation was
used (Pavelka 2001). When more than four control points were found and
implemented, the adjustment at individual point deviations from the ideal state
was found. Achieved deviations varied between 15 and 20 m on the ground control
points for the declassified satellite images from a relatively flat area (theoretical
geometrical resolution depends on the quality and satellite system; it can be 10 m or
better for high quality and resolution images). Better results cannot be achieved
without image orthogonalization. Due to the character of this project the absolute
accuracy is not the main factor. The main objective was to evaluate the content of
images on a case project that demonstrates the potential of this technology.
The main challenge was the image quality, which varies greatly. Images from
the 1970s have been worse from a radiometric point of view even after digital
image processing. In terms of evaluating the condition and type of vegetation
unambiguously these images are not suitable for such work because panchromatic range does not give a good alternative to separate classes using classification and the number of classes is very hard to recognize. Either additional
information about the landscape (e.g. archive maps) or multispectral images
should be used. Unfortunately these images are not available for older data.
The classification of high quality black and white aerial photographs into core
classes (various types of built-up areas, roads, fields, low vegetation outside the
agricultural fields, forests of different ages, isolated trees, water) achieved an
accuracy of 86–93 % class depended. Experiments were made using recognition
software based on object-oriented classification with help of newly calculated
channels (Halounova ´ 2004a, b, c).
Theoretically, it would be logical to start from the oldest images and to expand
the vector database to rectify based on visible changes. The procedure was necessary to reverse because the latest images are of the best quality. There is a sufficient
amount of additional material and the possibility of multispectral images for the
new images. It is also possible to assume that the character of some essential parts
hasn’t changed fundamentally. The final task was to choose a suitable interpretive
key from the Corine Land Cover.
48
K. Pavelka and E. Matous ˇkova ´