Research On SAR Image Segmentation Based On Statistical Information And Region Merging

Synthetic aperture radar(SAR)is a kind of radar that realizes microwave imaging with Doppler Effect.Because of its characteristics such as all-day,all-weather,high resolution,and strong penetration,SAR is widely used in crop estimation,monitoring of coastal erosion,detection of surface changes,tracing of enemy forces,precision strikes,and other civilian and military fields.However,since SAR is a slant projection type imaging,the speckle noise in the SAR image will make the SAR image suffer from strong interference in segmentation and interpretation.At the same time,the data quantity of SAR images is also increasing rapidly,which brings with it precise details and rich information,and increases the difficulty of segmentation techniques for SAR images.This paper focuses on the study of SAR image segmentation algorithm based on region merging,and proposes two SAR image segmentation algorithms.The first algorithm is based on perceptual hash and difference hash(PDHA),the second algorithm is based on ratio edge intensity map(RESM)and region merging(RESM_RM).The details of the two image segmentation algorithms are described as follow:(1)PDHA makes full use of the accuracy and robustness of perceptual hashing and the fast and efficient nature of difference hashing to achieve SAR image segmentation.In the preprocessing stage,enhanced Lee filtering is used to overcome the effect of noise on the segmentation of SAR images.Then,using the statistical characteristics of SAR images,multiple thresholds are selected for initial segmentation to obtain a set of homogenous regions according to the peak-valley values of the grayscale histogram.Then,the homogenous region set is used as an input for further segmentation,two different regions with greater similarity and exceeding a predetermined threshold are merged,and the region set is updated until the number of elements in the region set no longer changes.This algorithm not only improves the segmentation efficiency of the algorithm,but also improves the segmentation accuracy.Finally,experiments performed on SAR images illustrate the excellent performance of the proposed algorithm.(2)The RESM_RM algorithm uses a statistical information-based edge detection operator to replace the rotatable rectangular bilateral proportional edge detection operator.Firstly,a statistical information-based edge detector is used to capture the direction variable information of the SAR image,and a ratio edge intensity map(RESM)is constructed according to these direction variable information.Then watershed transform is used to obtain initialize segmentation with RESM,and the result of initialize segmentation is used as input in the next step.Finally,the over segmentation region in the initial segmentation is converted into a node in the region adjacency graph(RAG)and the fully connected graph(FCG).The line between the nodes represents the similarity between regions,and the region is gradually merged to get the final image segmentation results.Finally,experiments performed on SAR images illustrate the excellent performance of the proposed algorithm.