In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
459
4. DEVELOPMENT
Inputs such as images, digital topographic mapping and
software are needed to perform the work. The materials used
and the methodology are presented as follows.
4.1 Materials
The ALOS images and the topographical mapping comprising
the study area are shown in Figure 01. This area lies at Paraná
(Brazil) state, specifically in the municipality district of
Guaraque^aba, Brazil, near the coastline.
The area is bounded between the coordinates UTM:
• Upper left:
Latitude -25° 11'25.95"
Longitude -48° 22' 00.20"
• Bottom right:
Latitude: -25° 17'05.17"
Longitude -48° 14'37.92"
The ALOS images were acquired through the research program
of the Brazilian Institute of Geography and Statistics (IBGE),
which is responsible for distributing the ALOS images for
federal government agencies, research institutions and other
non-commercial users in Brazil. The software used is the ENVI
4.5 and MATLAB R2007b. The ENVI 4.5 allows the
manipulation of ALOS images as well as the topographic
mapping; the MATLAB is used for the development of routines
that take the mathematical morphology.
4.2 Methodology
The roads extracted were not pavement and belonging to PR -
405. To obtain the field reality, topographic maps at 1:25.000
scales were used to identify the roads on the scene. From the
step 01 to step 03, the processing was performed in the software
ENVI 4.5.
STEP 01 - REGISTRATION: The registration is required for
the integration of images acquired by different sensors, for
example the AVNIR-2 and PRISM, integration of images
obtained at different times, among other applications. For its
implementation is necessary to choose control points. After
collection of control points, which must be distributed across
the entire image, is needed to define the interpolation process.
In this work the bilinear interpolation method was chosen.
STEP 02 - FUSION: The images available are of different
sensors, in this case AVNIR-2 and PRISM and with different
spatial resolution, 10 meters and 2.5 meters respectively. Thus,
for the analysis of mixed spatial databases, which may consist
of images from different sensors with different spatial
resolutions, it is necessary to perform the fusion process. Fusion
was performed by principal components. After obtaining the
principal components, the first component is removed and
placed in the panchromatic image, in this case the PRISM
image. After PRISM assume to be the first principal component
is necessary to reverse transformation to return the multispectral
bands.
STEP 03 - CLASSIFICATION: This step is performed the
classification of the image obtained in the fusion process. The
method used was the supervised classification, where the
classes of information are pre-defined, and from its definition
are acquired samples of each class. To perform the
classification, classifiers are needed. The classifier used was the
statistical Maximum Likelihood - ML. The classified image is
used for verification of stretches of roads which are not possible
to identify with the PRISM image.
From the step 04, all processes have been developed in
MATLAB.
STEP 04 - EXTRACTION: An initial selection of stretches of
road is done using algorithms of mathematical morphology and
segmentation. For this process has generated a routine for the
PRISM image binarization. Armed with binary image the
morphological operators were applied to it. Operators of
closing, erosion were applied. Several tests using different
structuring elements were performed to find the best solution
for the image used. At this stage, most other classes, such as
vegetation, are eliminated. To eliminate other unwanted traces
of features the connected components of the image were
calculated and the smallest areas excluded. The product of this
post-processing was a cleaner image of traces of unwanted
features. However, at this moment the road obtained has not
complete linear features, performing inconsistently.
STEP 05 - FILLING: Was developed a routine to complement
the road obtained in extraction (step 04). This routine used
other techniques of mathematical morphology and an algorithm
for calculating the Euclidean distance. From a User-defined
threshold, based on the calculated Euclidean distance, the road
extracted was completed. At the end the images (PRISM binary
code, classified, filled) were overlaid to verify the areas filled by
the algorithm.
5. RESULTS
The methodology has been applied and the results were as
follows. Initially the PRISM image panchromatic was binary
code in MATLAB and the road cut into two parts. For this
result was used a threshold equal to 90. This threshold is
defined as object to be extracted in the image. After the
application of morphological operators like dilation, was not