Remote Sensing Image Registration Based On Multi-Features

With the development of Aeronautic technology,taking images by satellite andairplanes has already become the important means to obtain the information of ground.Remote Image has the characterristic of larger area and richer contents. ect. The research ofthis paper studies best matching two or more images at space place from the same area,where the images taken at different visual angles or taken by different time space ,which isbased on remote image registration.For image matching,feature extraction is the very important part. This papercombines the geometric feature with feature based on edge and puts forward a kind ofmatching method based on multi-features。This method combines the usage of affinemoment invariants and Edge gradient. Affine moment invariants reflects geometric featureof image and can be obtained via algebra invariance theory .It is not sensitive to rotation,scale,translation transformation. Considering the characteristic that Remote sensing imagehas large amounts of information, it may be impossible to distinguish similar area onlydepending on image geometric feature. Edge features is the very important characteristic ofimage and can be get easily. There already have been many edge-extraction algorithms, suchas Sobel operator,canny operator,Log operator etc. This paper takes the edge extractionmethod based on Curved surface fitting. First do Curved surface fitting using cubic spline,and then take the edge using parameter of cubic spline. Find the maximum and minimum ofEdge gradient and the maximum value of forstner operator from image edge,and regardthem as image features. parameter of cubic spline has better ability of imagenoise .Combining affine moment invariants,maximum and minimum of Edge gradient,Forstner operator described above mean value,variance, they together constitute the imagematching characteristic.In the matching algorithm, set a difference of eigenvalue bound value for each feature,and then make subtraction of eigenvalue drawn from two different images. If the absolutevalue is lower than bound value, then comparability increasesd, otherwise comparabilitydecreased. compare all eigenvalue to get the last comparability. In the process of computingcomparability ,also can set weight value according to real world for each characteristic toreflect the importance of corresponding characteristic for image matching.