Segmentation-based Change Detection Method for Remotely Sensed Images
2. METHODOLOGY
2.1 Multi-dated composite image
First of all, the remote sensing images acquired at the same location but different dates are
grouped together to create a multi-dated composite image. In order to put the image
composition into effect, an assumption has to be made that both geometric and spectral
disparities between multi-dated images have been properly corrected beforehand. By doing so,
a multi-dated composite image will be produced for detecting the changed areas that will be
described in the next subsection.
2.2 Region growing segmentation
This study uses region growing approach to perform the image segmentation. Region growing
is a procedure that groups the image pixels with similar gray values into regions. It firstly
chooses a seed pixel and makes the gray value comparisons with its neighbor pixels. A region
will therefore start and grow when the neighbor pixels have gray values similar to the seed pixel
and stop when the gray value similarity no more exists. Then another region will begin to
grow with new seed. The growing process will continue until all of the pixels have been
assigned to their belonging regions.
2.3 Change detection by statistical significance test
The changed or no-changed condition of the regions generated from previous region growing
process will be analyzed here. In this study, the gray values of the multi-dated region are
statistically analyzed using Chi-square test. The significance of the test is using Chi-square
values to test the degree of similarity between two sets of data. Therefore, Chi-square value of
each region is calculated for multi-dated images, and is tested for their statistical significance by
null hypothesis. The region will be considered as a changed region when the chi-square value
is greater than the threshold value under certain confidence level.
