Adaptive Superpixel Generation forPolarimetric SAR Images with Local IterativeClustering and SIRV Model
(English)In: IEEE Transactions on Geoscience and Remote Sensing, ISSN 0196-2892, E-ISSN 1558-0644Article in journal (Refereed) Submitted
Simple linear iterative clustering (SLIC) algorithmwas proposed for optical images and shows satisfactoryperformance on superpixel generation. Several studies havebeen proposed to modify SLIC to make it applicable for PolSARimages, where the Wishart distance is adopted as the similaritymeasure. However, the superpixel segmentation results of thesemethods are not very acceptable in heterogeneous urban areas.Further, the trade-off factor which controls the relative weightbetween polarimetric similarity and spatial proximity is not easyto be determined. An adaptive polarimetric SLIC (Pol-ASLIC)superpixel generation method is proposed in this paper toovercome these limitations. First, the spherically invariantrandom vector (SIRV) product model is adopted to estimate thenormalized covariance matrix and texture for each pixel. A newedge detector is then utilized to extract PolSAR image edges forcentral seeds initialization. In the local iterative clustering,multiple cues including polarimetric, texture and spatialinformation are considered to define the similarity measure.Moreover, a polarimetric homogeneity measurement is used toautomatically determine the trade-off factor, which can vary indifferent homogeneous and heterogeneous areas. Finally, theSLIC superpixel generation scheme is introduced to the PolSARdata. This proposed algorithm produces compact superpixelswhich can well adhere to image boundaries in both natural andurban areas. The detail information in heterogeneous areas canbe well preserved. Airborne ESAR and PiSAR L-band PolSARdata are used to demonstrate the effectiveness of this proposedsuperpixel generation approach.
Place, publisher, year, edition, pages
Spherically invariant random vector (SIRV), superpixel, edge detection, Simple linear iterative clustering (SLIC), polarimetric SAR (PolSAR).
IdentifiersURN: urn:nbn:se:kth:diva-187944OAI: oai:DiVA.org:kth-187944DiVA: diva2:932590
QC 201606072016-06-012016-06-012016-06-07Bibliographically approved