Change Detection Method Based On Cosegmentation For Remotely Sensed Image

Land cover is the synthesis of various material type of earth surface and its natural attributes.With the development of aeronautics and astronautics technology,earth observation of satellite covering full range of globe has been completed.The remotely sensed satellite data can be used to acquire land cover information of a large area or the whole globe,which helps timely and accurate obtain the distribution and change of earth's surface cover types.The successful development of land cover remote sensing products greatly facilitates the application of land cover information in the field of climate change research,ecological environment monitoring and sustainable development planning.In order to meet the needs of land cover for present situation,a number of methods has been developed to detect change information from multi-temporal images for the aim of updating land cover products.The core of change detection method is how to detect change,determine range and attribute of change.Several change detection methods take image pixel as the basic processing unit and use the spectral value to construct the feature,the threshold is the criterion for judging change or non-change.In this case,it is inclined to obtain false changes or omission error,even if the combination of multiple feature is used.Object-oriented change detection method take spatial information of the image into consideration by treating the homogeneous object as the basic unit.The optimum segmenting scale for different land cover types and the correspondence of multi-temporal images objects remain to be hot issue.However,the threshold is still the key factor that affects the accuracy of change detection.The cosegmentation algorithm in computer vision can segment the same or similar object from the multi view image of the same scene.The algorithm can extract more image information because it takes advantage of the connection among the image group.If the change process of land cover is regarded as the movement of nature,the change detection of land cover can be regarded as the cosegmentation of motion image.Considering the pixel-based and object-oriented change detection methods,which taking the threshold as the criterion to judge the change / non change,in this paper,an energy function suitable for change detection is constructed according to the idea of cosegmentation.Based on the common change information of two phase image,the change feature term in energy function uses the threshold condition as a priori knowledge to judge whether the change occurs.At the same time,the image feature term included in energy function can reflect the similarity between pixels in the neighborhood in order to make better use of the spatial information of the image.In this paper,graph cut-based method is used to solve the minimization problem of energy function.To extract corresponding change patch from multi-temporal images,images will be mapped to graph and the terms of energy function will be used as weights of edges in the graph in the process of min-cut/max-flow method.The content of this paper is arranged as follows:The innovation point of this paper are as follows: 1.According to the common characteristic of change detection method and cosegmentation,the threshold condition of change information will be used as priori knowledge of change feature term in energy function,while change detection method of pixel-based and object-oriented consider it as the standard of change and non-change.2.The construction of image feature term in the energy function takes spectral and text similarity into consideration of the pixel pairs in the neighborhood.By giving different weight values of multi-temporal image feature terms,a comprehensive image feature item is constructed.According to the segmentation results of different image features,it can not only extract the location and region,but also measure the geometric properties.