Change Detection In SAR Images Based On Structural Information And Saliency Detection

Change detection is an important technology in the field of image processing,it analyzes multi-temporal remote sensing images acquired on the same geographical areas in order to identify changes occurred and locate change regions.Synthetic Aperture Radar(SAR),as the main technical means of remote sensing imaging,has unique advantages in change detection because it is insensitive to illumination and atmospheric conditions and its capability of imaging in all-weather condition day and night.Recently,change detection based on SAR images has been used successfully in many applications,such as environmental protection,agricultural development,analysis of urban changes,earthquake Relief,and military reconnaissance.However,due to special sensor characteristics of SAR and the presence of coherent speckle noise during the acquisition of SAR images,there are two issues devoted to the change detection based on SAR images: 1)dealing with multiplicative speckle noise;2)maintaining edge and structural details.For that,the main research in this paper is to take the two problems into account.For change detection,structural information is of central importance for maintaining edge and textures.Extracting structural information into change detection helps maintain detailed information in the image.The phase consistency can detect the structural features of images in all directions,and thus can extract rich and accurate texture structure information,and produce edge features consistent with the human visual system,as well as is not affected by brightness and contrast of images.Thus,in this paper,we use phase consistency extract structural information to solve the problems of edge location and structure retention in change detection.In addition,for change detection,significant detection can extract the most interesting areas and targets in images.So,there are two major roles of saliency detection for change detection: extracting changed regions and eliminating the noise interference in background regions.Therefore,we introduce the saliency-based sub-regional fusion strategy for change detection to maintain detailed information and suppress noise.As the above discussion,we propose a SAR image change detection method based on structural information and saliency detection,the main works of this paper are as follows.1.A SAR image change detection method based on phase congruence and neighborhoodbased ratio(PCNR)is proposed.The method uses the phase congruence technique based on non-local mean to extract the structural information in the image and uses the structural information as the weight parameter to measure the heterogeneity of the difference image.In the homogeneity region with lower heterogeneity parameter,the mean ratio is used to calculate the difference image,while in the heterogeneous region,the ratio operation is preferred,and the weight parameter is used to combine the mean ratio and the ratio operation.So the obtained difference image maintains the edges and structural details of images as well as suppress noise.Finally,the Otsu algorithm is used to segment the obtained difference image to obtain the final change detection map.The availability and feasibility of the proposed method are verified by conducting experiments on the real SAR images.2.A SAR image change detection method based on saliency sub-region and wavelet decomposition is proposed.The core of the method is to use the saliency map obtained by saliency detection as the guidance of sub-region multi-scale fusion.First of all,the stationary wavelet decomposition is performed on the log-ratio difference image and the PCNR difference image,and the retained low-frequency components according to the difference images are also different.Then,the retained multi-scale low-frequency components are segmented by the FCM clustering and Bayesian classification algorithm,and the change detection maps corresponding to each scale are obtained.Finally,the multi-scale fusion strategy,based on the saliency map,is used to fuse the multi-scale change detection maps,and the final change detection map is obtained.In this method,from the acquisition of the difference map to the wavelet decomposition to the multi-scale fusion based on the salient map,the suppression of the speckle noise as well as the preservation of edge information and texture textures are considered.In the analysis and comparison of experiments,the method can maintain the details better,and has stronger ability of noise suppression and higher detection accuracy.