Feature Based Automatic Registration Of Remote Sensing Images And Remote Sensing Data Processing System

Image registration is a very essential and important step in many research directions of remote sensing image processing,because it is the foundation basis of many subsequent processing steps such as image analysis,computation,comparison,fusion and so on.For remote sensing,image registration refers to the best match between two or more images obtained from different phases,different viewing angles(shooting position and angle),different imaging devices or different natural conditions(sunlight or weather,etc.),and then conduct the process of superposition.During the registration process of homogeneous remote sensing images,there are some problems such as transformation of illumination and weather conditions,deformation of objects,displacement and so on,that will affect the registration accuracy.As the optical remote sensing image is sensitive to natural conditions such as sunlight and weather,the quality of image by sensor will be affected by lacking illumination and also cloud and mist.It may lead the remote sensing data blurring or missing.In addition,the earth's surface will alter significantly when encounter large-scale natural disasters or great area fire,explosions and other man-made disasters.These situations will inevitably affect the selection of counterpart points,which ultimately lead to inaccuracy of the registration results.According to the above problems,this paper mainly focuses on the study of automatic registration algorithm among homologous remote sensing images.In addition,this paper has developed a human-computer interaction software platform for multi-spectral remote sensing image processing,which relies on the project funded by central university basic scientific research business special funds.The platform can better realize the large data file management,and also the remote sensing image processing and analysis.The remote sensing images can be interpreted well through the system.The specific work is summarized as follows:1.For the deficiency in the accuracy of registration when SIFT(Scale Invariant Feature Transform)feature applied to the situation that cloud and mist shielding or deformation of objects,this paper proposes a new SIFT variant,GR-SIFT(Scale-invariant Feature Transform based Gradient),which based on the original theory of SIFT algorithm,by improving the traditional gradient calculation method.The method is implemented by introducing an exponential weighted average filter in the gradient calculation,and a new descriptor is generated by combining the logarithmic polar descriptor.Experimental results show that this method can obtain higher registration accuracy for multi-spectral remote sensing images,and achieves the sub-pixel application requirements.2.The software system developed in this paper is mainly aimed at analyzing the spectral remote sensing image,which against to construct a human-computer interaction system for remote sensing image processing.It can integrate image display,contrast stretch,ground targets classification,road network detection,automatic registration and other functional modules,and also create a reserved interface that can be achieved with the follow-up function,which expected to achieve variety functional expansion.