Research On Feature-based Registration And Fusion Of Optical And SAR Images

In order to maximize the availability of image information,heterogeneous remote sensing image fusion integrates data with complementary advantages,which has important practical value in the fields of medical treatment,remote sensing,and aerospace.The advantage of optical imaging is that it has rich colors and high spatial resolution,and can clearly observe the rich texture details of ground objects.The advantage of SAR is that it is active radar,which is not disturbed by bad weather for 24 hours and has a certain degree of penetration.Registration is a necessary prerequisite for image fusion,so it is of great significance to research the registration and fusion of optical and SAR images.At present,with the development of remote sensing technology,optical and SAR images have significantly improved their spatial resolution,and traditional registration methods have failed.In terms of pixel-level fusion,optical and SAR images have a large difference,and SAR has a narrow gray-scale range,which results in the structure in the SAR image cannot be smoothly injected into the optical image,which causes the problem of spectral distortion in the fusion result.In view of the above problems,this article improves the method.The main research contents are as follows:(1)For feature point detection,in order to detect more stable and reliable feature points,this paper proposes to construct a nonlinear scale space of anisotropic diffusion,and then in this space feature point extraction is performed on each layer of image.The filter kernel used in this scale space is nonlinear,and the Gaussian kernel used by the Gaussian pyramid filter kernel is the only linear filter kernel.The filter kernel in the nonlinear scale space overcomes the disadvantage that the Gaussian kernel will blur the image.A spatial scale image is clearer.(2)In order to resist the non-linear radiation difference caused by the imaging mechanism of heterogeneous images,this paper uses phase consistency to project the two images onto a more consistent space.In terms of feature vector construction,since most of the pixel values in the phase consistency value(PC)mapping are close to zero,the information of the PC mapping is less and insufficient for feature description.Secondly,the PC map is very sensitive to noise,because it mainly contains edges,which will lead to inaccurate feature descriptions.In response to the above problems,this article uses the maximum index map based on PC information instead of directly using the PC map to describe.(3)Since only the intensity value(PC)is used to describe the features,it is still not very robust.Therefore,this paper proposes a comprehensive use of phase consistency information to increase the direction attribute of phase consistency and increase the degree of discrimination between similar feature vectors and dissimilar feature vectors.(4)In the fusion problem,the structure in the SAR image cannot be smoothly transitioned when injected into the optics,which causes the problem of spectral distortion in the fused image.This paper proposes a fusion algorithm based on NSST transform combined with sub-regions.This algorithm mainly replaces the IHS inverse transform with a sub-region method to distinguish between SAR target regions and non-SAR target regions,and then two different fusion rules are designed for these two regions.