Object-Based Urban Change Detection Using High Resolution SAR Images

In this study, the unsupervised detection of urban changes, based on high-spatial resolution SAR imagery, is approached using the object-oriented paradigm. Multidate images segmentation strategy was adopted to avoid the creation of sliver polygon. Following segmentation, a change image was generated by comparing objects' mean intensities using a modified version of the traditional ratio operator. Three different unsupervised thresholding algorithms-that is, Kittler-Illingworth algorithm, Otsu method, and outlier detection technique-are used to threshold the change image and generate a binary change map. Two TerraSAR-X SAR images acquired over Shanghai in August, 2008, and September, 2011, were used to test the methods. The results indicate that, compared with pixel-based, the obj ect-based approach helps in improving the quality of the produced change maps. The results also show that the three unsupervised thresholding algorithms performed equally well.