| Due to its all-weather day-and-night imaging capabilities,Synthetic Aperture Radar(SAR)has been widely used in the fields,such as dynamic monitoring of sea area and automatic navigation etc.In sea area management,it is an essential solution for coastline detection to monitor dynamic changes of the coastline.However,due to the strong speckle noise in SAR imagery,the variety of sea states,and the complexity of coastal environment,it is very difficult to extract accurate coastline.From the perspective of superpixels,this paper mainly studies the coastline detection algorithm under the condition that some dissimilar objects exist in land and ocean with low contrast in SAR images.The main works of this paper are as follows:(1)A modified coastline detection algorithm combining region merging and superpixels is proposed.When the ocean surface or land is heterogeneous,the existing region merging based coastline detection algorithm is hard to merge the small area,and needs to set the threshold artificially.To solve this problem,firstly,an improved superpixels algorithm is proposed,in which a modified local window can effectively be utilized to solve the problem of fuzzy feature in contrast with the traditional rectangular window.Secondly a local similarity descriptor is constructed,to obtain good edge fitting degree.Finally,an improved regional merging criteria is given based on superpixels as a basic element,from which the pixel mean,the relative size and statistic information of superpixels are taken into account,and local threshold value,which has to be set artificially in existing algorithm,can be calculated according to the neighborhood information.Experimental results show that the proposed algorithm is effective on real SAR images.(2)A modified coastline detection algorithm combining superpixels and Three Markov Random Field(TMF)is proposed.In order to solve low edge fitting degree of existing algorithms,firstly,a super-pixel algorithm based on Gamma distribution is proposed.It is assumed that all pixels,in a homogeneous local neighborhood,obey Gamma distribution,from which left and right thresholds are calculated by the confidence interval to form superpixels so as to improve the edge fitting degree.Secondly,an improved superpixels based TMF algorithm is presented,in which an energy function based on the improved auxiliary field is constructed to solve the threshold dependence problem in the existing TMF algorithm.Experiments results indicate that the proposed algorithm has a good performance on real SAR images.(3)A modified coastline detection algorithm combining superpixels and probabilistic Three Markov Random Field is proposed.Firstly,a statistics is proposed to enhance the contrast between sea and land.Secondly,an improved superpixels algorithm is given to solve the problem of low edge fitting degree.Thirdly,an improved probabilistic TMF algorithm is proposed which takes into account the correlation among superpixels.A probabilistic vector is utilized to represent the probability that the current superpixel belongs to sea or land.Finally,this vector is incorporated in the energy function based on probability to replace Ising model in the existing TMF.The effectiveness of the algorithm is verified by the experiments on real SAR images. |
PDF Full Text Request