Study On PDE Methods For InSAS Interferogram Filtering

Interferometric Synthetic Aperture Sonar (InSAS) provides a means of obtaininghigh resolution three dimensional images of targets on the sea floor. It increases one ormore receivers based on Synthetic Aperture Sonar(SAS). Three-dimensionalinformation of the objects is derived from the phase difference between two sonarimages taken from separate receiving arrays. InSAS, as one of the new high-resolutionimaging technologies underwater, has many potential applications in both military andcivil fields, such as mine hunting and seabed topography mapping.Because of its intrinsic coherence-imaging method, the interferometric phaseimage obtained directly from the InSAS images often exhibits much noise, which willaffect the following phase unwrapping result and the accuracy of Digital ElevationModel(DEM). Therefore, denoising is always an important research in InSAS imageprocessing. In recent years, Partial Differential Equation(PDE) has been widely appliedin image processing field, such as image smoothing, image segmentation, edgedetection and so on. In this thesis, supported by National High Technology Research andDevelopment Program, the PDE method in InSAS interferogram processing is studied.The aim of filtering is to effectively save fringe and texture information of image whenremoving the noise.This thesis, at first, introduces the principle of InSAS imaging, characteristics ofinterferogram and the sources and type of noise. Then it gives some kinds of evaluationcriterion and proposes an new evaluation criterion, that is edge correlation coefficient. Then it presents some general filtering methods and analyzes their advantage anddisadvantage. This thesis mainly focuses on PDE methods, such as P-M model,anisotropic diffusion model and coupled PDE model. It analyzes the edge behaviours ofP-M model and its disadvantages, and proposes the regularization method of P-M model.The numerical implementation and algorithm steps for those three models are presentedin this thesis. Using simulated and real data generated from InSAS images of QianDaolake in Zhejiang Province, the results of experiments demonstrated that, compared withaverage filter and midain filter, PDE method can not only denoise more effectively, butalso preserve edge and texture information of interferogram. In the last of this thesis,the application of PDE method on rock image processing is introduced briefly.