Segmentation of Polarimetric SAR Image Using Object-Oriented Strategy

In this paper, a hierarchical segmentation algorithm using object-oriented strategy is proposed to partition polarimetric SAR data into homogeneous regions efficiently. After the multi-looked and de-speckled processing, the scheme uses a homogeneity definition to achieve an optimal result in local and global through a hierarchical stepwise process. Since PolSAR data contains plenty of polarimetry information, the neighboring regions in an image are partitioned if they are different in polarimetric behavior. Then the homogeneity is obtained by utilizing the decomposition components of PolSAR and shape features, which are under the object-oriented strategy. Under different decomposition theory, the polarimetric heterogeneity are different, thus the two segmentation criteria proposed vary with each other. To validate the object-oriented scheme, ESAR L-band PolSAR data is used for testing. Experimental results show that our model works well. And results from different criteria demonstrate that it is better to choose proper decomposition method for given targets.