Interactive Extraction Typical Objects From Very High Resolution Remote Sensing Images

With the rapid development of acquisition technology of very high resolution(VHR)remote sensing image,the development of remote sensing images in the direction of high space,hyperspectral,and high temporal resolution makes the features such as color,texture,and shape of the image more prominent.Therefore,it is possible to accurately extracting objects using VHR remote sensing images.However,due to the complexity of remote sensing images,automatic interpretation of computer can not meet the accuracy requirements.In practical applications,it mostly depends on visual interpretation of human,which requires a lot of manpower and material resources and is inefficient.With the large amount of data from remote sensing,artificial visual methods have become seriously inadequate.Based on the above considerations,the interactive extraction method as an alternative method,combines full people's recognition ability and computer's processing ability,which not only guarantees the understanding of translation accuracy,but also improves the understanding of translation efficiency.This paper carries out the research on the interactive extraction method of right-angle buildings and natural objects from VHR remote sensing images,respectively.The research contents are as follows:(1)For the right-angle buildings with a star shape,this paper propose a method of interactively extracting right-angle buildings.Firstly,the image block containing the target building is obtained by manual interaction.Next,the image block is preprocessed by bilateral filtering.Then the graph cuts with the star shape constraint is used to obtain the building objects.Finally,building object is regularized into real regular shape through corner detection and linear fitting.The experiments performed on two different region and spatial resolution aerial imageries demonstrate the stability and accuracy of the proposed method.(2)For the water,woodland,terraced fields,and bare ground,called natural objects,with abundant color and texture information,this paper proposes a method of interactively extracting nature objects based on fully connected conditional random fields.Firstly,the input image is segmented to obtain superpixels with similar characteristics.Then the foreground samples are marked by human interaction.Next,the color and texture features of each superpixel are extracted.Then a fully-connected conditional random field model is established to combine the user interaction markers.Finally,using the mean-field estimation supported by the highdimensional Gaussian filtering to achieve model inference,the target contour is obtained.By extracting a variety of ground features from a high-resolution remote sensing image,the experimental results prove the effectiveness of the proposed method.