A Sparse Triplet Markov Fields Model For SAR Image Segmentation

Synthetic aperture radar(SAR)is a kind of active aerospace and aviation microwave imaging system,and it has the all-time,all-weather,multiband,and Multi-View abilities to acquire ground observation data,which can be widely used in various military and civil applications.SAR image segmentation is a precondition that the SAR image can be effectively analytical,segmentation can effectively reflect the spatial correlation of target and reveals the essence of SAR image,so it is widely studied.Because the SAR image is often affected by the multiplicative speckle noise,this affect the image quality seriously.The emphasis and difficulty of the SAR image is that how to suppress speckle noise and at the same time segment the SAR image accurately.Markov random field(MRF)model can describe the pixels correlation which in neighborhood,therefore this model is widely used in image processing field.Because the built model is based on the pixels isotropic in neighborhood system,so the model cannot describe the non-stationary SAR images which have complex texture structure.Triplet Markov fields(TMF)model introduced auxiliary field U and depict the non-stationary SAR image,the segmentation result is satisfactory.However,the TMF model captured the feature between pixels in low order neighborhood is very limited,and cannot fully describe the edge and outline of the target.So when segmenting the SAR image which have complex scenarios and abundant targets,the segmentation result is not satisfactory.Based on the TMF model,in this paper,we propose a sparse triplet Markov field model for SAR image segmentation algorithms.The method breaks through the restrict that the lower order neighborhood TMF mode cannot accurate describe the complex SAR image,excavate the pixels otherness which belong to the different classes in high order neighborhood,and improve the SAR image segmentation result.Under the STMF model framework,first of all,this model suggests using a higher order neighborhood,redefined the auxiliary field U based on adaptive non-stationary division,complete the pixel~'s similarity division of the higher order neighborhood and determine the boundary of the two unrelated pixels,by using the auxiliary field guide the boundary of the SAR image positioning.Secondly,build the sparse potential energy function which based on higher order neighborhood,compute the reconstruction error of the related pixels,will the error as sparse energy and is applied to the potential energy function.Because the priori restrictions was imposed on the pixel,so the segmentation result accuracy of SAR image is improved.Last,the parameter is estimated by the iterative condition model(ICM)algorithm.By the real SAR image simulation,the results show that the algorithm can restrain multiplicative speckle noise interference,improve homogeneous regions'consistency in SAR image,the local heterogeneous complex region also get accurate segmentation,and effectively improve the positioning precision of the segmentation boundary,the proposed algorithm is superior to the existing of MRF and TMF segmentation algorithm.