| Synthetic aperture radar(SAR)imaging has unique advantages in sensing the scene information and target characteristics.Due to the independency of the weather and sunlight condition,the strong detection ability of SAR system in large scenes plays an important role in civil and military fields like target detection and recognition,battlefield analysis,landform prospecting,and disaster assessment.As the imaging technique and quality increasingly improve,the more and more high-quality SAR images thus tend to require the refined,automatic,and intelligent interpretation of SAR images in applications,and the existing methods encounter some technical bottleneck.In terms of the insufficiency of traditional methods in robustness and refinement,the robust edge detection and refined scene segmentation are researched in SAR images of complex scenes.The main research contributions in this dissertation are summarized as follows.1.SAR image edge detection robust to isolated strong scatterers using anisotropic morphological directional ratio test.Edge detection via the traditional anisotropic average directional ratio(AADR)for synthetic aperture radar(SAR)images produces unwanted and fragmentized edge pixels due to the speckle noise and sporadic highlight pixels from isolated strong scatterers in real scenes.Thus,a novel robust edge detector is developed for SAR images.The anisotropic morphological directional ratio(AMDR)is proposed to describe the intensity variation in SAR images by combining rotated biwindows and the weighted median filtering(WMF),and the response of AMDR to the ideal step edge is deduced as well as the width of the response and the edge resolution constant.Then,multiplicative spatial and directional matched filters are presented to improve edge localization and direction estimation of the AMDR.Based on the improved AMDR,the edge strength map(ESM)and the edge direction map(EDM)are extracted and embedded into the route of Canny detector to extract thin edges.Moreover,an edge remedy processing is given to improve the connectivity of edges.Finally,performance comparison is conducted with the existing AADR-based detectors on synthetic and real SAR images using the receiver-operatingcharacteristic(ROC)curves,which shows that the proposed AMDR-based detector evidently possesses more robust ability to counteract sporadic isolated highlight pixels in SAR images than the existing AADR-based detectors do.Moreover,in the case without sporadic isolated highlight pixels,its performance is also comparable with those of existing competitive AADR-based ones.2.A segmentation algorithm for SAR images via region merging guided by morphological edge information.In synthetic aperture radar(SAR)image segmentation based on region merging,the over-fragmentation of initial partition affects the efficiency and quality of subsequent region merging.In order to solve this problem,a segmentation algorithm is proposed for SAR images using region merging guided by morphological edge information.An anisotropic morphological directional ratio operator is firstly constructed by using the weighted median filter that is weighted by the Gaussian-Gamma function to extract the edge strength map(ESM)of an image as the strength of edge responses.Then,the thresholding processing and watershed transform are performed on the ESM to obtain a highquality initial partition.During the region merging process,the pair of adjacent regions with the smallest merging cost are iteratively merged by using the technique of relative common boundary length penalty,and the final segmentation is outputted until the termination condition of region merging is satisfied.Experimental results on real SAR images show that the proposed method evidently improves the quality of initial partition,and the number of regions in the initial partition reduces by over 25% under the premise of ensuring the final segmentation quality,which accelerates the region merging process.Moreover,the final segmentation results of the proposed method are generally better than those of the compared ones based on region merging in different indexes of performance evaluation.3.SAR image edge detection via directional Bhattacharyya coefficient.The robust edge detection for SAR images,a novel edge detector is proposed by introducing the Bhattacharyya coefficient(BC)combining with the rotated biwindow configuration.The scenes in SAR images are complicated and the edge detection statistics based on traditional low-order statistics like the mean possesses a limited ability to describe them while the grayscale histogram can better describes the statistical characteristics of regions than the low-order statistics.Therefore,based on the quantified input SAR image,the BC is computed between two normalized grayscale distribution histograms of local regions supported by the biwindow on the opposite sides of the pixel to be detected.With biwindows of different directions sliding through the image,multiple directional Bhattacharyya coefficients are obtained,which are utilized to extract the edge strength map(ESM)and edge directional map(EDM)of input SAR images.Embedding them into the framework of Canny detector can yield the final result of edge detection.Experimental results on real SAR images show that the proposed edge detector can accurately localize edge pixels of textured regions and effectively detect weak edges.Moreover,the BC-based ESM can be used as a good precursor to guide SAR image segmentation based on region merging.4.Region merging method with texture pattern attention for SAR image segmentation.In order to improve the accuracy of segmenting textured regions in SAR images,a region merging method is proposed for SAR image segmentation based on the texture pattern similarity.The proposed segmentation algorithm mainly incorporates two stages.Firstly,by the initial partition method based on the multiscale Bhattacharyya distance,an input image is over-segmented into considerable region patches with fine boundaries,which ensures the similarity of pixels in each patch.The second stage is the mandatory and selective merging process.For small fragmentized regions,they are merged into the most similar adjacent larger region by using a mean-based merging measure,which ensures the sufficiency of samples for the sequent construction of the texture similarity measure of regions.The selective merging process utilizes the spatial correlation matrix to construct the texture pattern similarity measure(TPSM)between adjacent patches,which is fused to obtain the proposed merging cost with a statistical similarity measure(SSM)and relative common boundary length penalty(RCBLP)term.Under the region merging framework of undirected region adjacent graph(RAG),the pair of adjacent regions with the smallest merging cost in the first-stage segmentation are iteratively merged,and the final segmentation result is output until all the rest merging costs are larger than a predefined threshold.Experimental results on real SAR images indicate that measures in different stages make the proposed method possess high accuracy in segmenting SAR images of complex scenes,and the proposed method outperforms several compared state-of-the-art ones based on region merging. |
PDF Full Text Request