Remote Sensing Image Fusion Based On Multi-scale Transform

Multi-source remote sensing data makes a great contribution to the scientific research and development of remote sensing image fusion.The images obtained by different imaging principles focus on different information,and the information that can be displayed by a single-source image is relatively limited.Therefore,under the same geographic coordinates,the multi-source remote sensing image information obtained at the same time is fused.For example,the images with the high spatial resolution and colorful information can be obtained during the process.In order to overcome various troubles that the design of the pansharpening algorithms bring into the fused images,this paper proposes the following two pansharpening algorithms.Aiming at overcoming the lack of spectral information richness in current pansharpening algorithms,this paper designs a fusion algorithm based on relative total variation structure extraction.Firstly,to extract the structure from the source image,the relative total variation method is used,at the same time,the difference information is used as texture,corresponding to the base bands and detail bands of the source images,respectively.Secondly,a visual decision graph is introduced in the high-frequency layer to guide the high-frequency fusion.In the base layer fusion rule,the edge strength of the panchromatic image is considered as the key character of the images,and the color information strength of the multispectral image is also nonnegligible during the process.Lastly,the fusion image is obtained by reconstructing the low-frequency fusion result and the high-frequency fusion result.The algorithm fully extracts the texture of the image through the relative total variation structure and retains it during the fusion process.The design of the fusion rule focuses on the spectral information of the multispectral image which has better effects compared with other similar algorithms.Traditional remote sensing image fusion algorithms are prone to problems such as the loss of detail information,artifacts and spectral distortion.To solve the above problems,this paper presents a pansharpening algorithm based on real-time image smoothing ILS and image similarity.Firstly,the multispectral image is converted by IHS,and the iterative least squares method is decomposed in real time for its I component and panchromatic image.Secondly,in the high-frequency fusion,the image similarity is designed to guide the high-frequency self-contained fusion;in the lowfrequency fusion process,the image saliency is used to guide the image fusion.Finally,the fusion result is reconstructed from the fused I component,as well as the H and S components through inverse transformation,and the extracted difference information is injected into the final fused image.The advantages of the pansharpening algorithm in this paper are that it overcomes the problem of color distortion caused by the traditional transform domain methods,and designs the image similarity to preserve the details of the source images better.The algorithm in this paper has got a great improvement in final effects and has certain advantages during the comparative experiments.