Feature-based Optical And SAR Image Registration Method

Optical remote sensing and SAR remote sensing,as two important earth observation technologies,play an irreplaceable role in military,emergency rescue and other fields.However,both have certain limitations.Therefore,it is necessary to comprehensively utilize optical and SAR images.The registration technology of optics and SAR images is the prerequisite for their comprehensive utilization,and has gradually become a research hotspot in remote sensing image processing.Optical remote sensing relies on reflected sunlight for imaging,while SAR remote sensing relies on receiving and transmitting echoes for imaging.This difference in imaging mechanism causes nonlinear intensity differences between optical and SAR images.At the same time,SAR images are also affected by speckle noise.This makes it difficult to extract feature points with a high repetition rate in the two images,and it is difficult to construct feature descriptors with modal invariance,which brings great challenges to the registration.This paper proposes solutions to some problems in the registration of optics and SAR images:(1)Aiming at the problem that it is difficult to extract high repetition rate feature points in optical and SAR images,this paper uses multi-scale Sobel operator and multi-scale ROEWA(Ratio of Exponentially Weighted Averages)operator to calculate the consistency gradient for optical and SAR images respectively.Among them,the ROEWA operator can resist the speckle noise in the SAR image and reduce the false feature points formed by the noise.On this basis,the scale space is constructed,and stable corner features can be extracted.(2)Aiming at the problem that it is difficult to construct feature descriptors with modal invariance in optical and SAR images,this paper uses MIM constructed based on phase consistency as the feature descriptor.Because there is a nonlinear intensity difference between optical and SAR images,but the structural information in their images remains similar,the phase consistency is used to extract the structural information in the image,and the structural information is used to construct the descriptor MIM.In order to make the corresponding point feature descriptors better match,this paper adds point position,scale and main direction constraints to make the matching result more accurate,reduce mismatches,and effectively improve the registration accuracy.