A Pan-sharpening Method Based On Spatial Information Enhancement

The Pan-sharpening is to obtain the multispectral image of high spatial resolution,high spectral resolution.In recent years,researchers have study the pan-sharpening,which can be get multispectral images with high spatial resolution and high spectral resolution.At present,there are lots of work on the Pan-sharpening,but there are some problems: on the one hand,the traditional method of injecting the details of the characteristics of redundancy,which will lead to color distortion of the fused image.On the other hand,the image is a set of pixels with spatial relations,but the spatial relations are not considered in the traditional Pan-sharpening.In this paper,we focus on the problem of enhancing the spatial information of the pan-sharpening method.Specific contents are as follows:(1)A new pan-sharpening based on Spatial Probability PCA is designed.In order to better understand the relationship between multi spectral images,this paper introduces the spatial probability PCA to eliminate the assumption that the PCA transform implies the assumption of independent and identically distributed.The super pixel segmentation algorithm from multispectral image on a structure similar to that of the block,to construct these similar structure block space relation matrix and Euclidean distance is used as a direct replacement;panchromatic image matching,image fusion of redundant features of the panchromatic image,we use the NSCT decomposition of PRPCA extract the first principal component of the panchromatic image and the high the low frequency part,there is a certain relationship between the coefficient and the coefficient of around we are now in the experiment,we further propose a new fusion rule based on local energy.The experimental results show that the proposed method performs better than the traditional method for panchromatic image algorithm in the traditional method in the ERGAS index with 0.1-0.5 improvement.(2)A new algorithm for pan-sharpening based on multi-scale low rank decomposition is proposed.Considering the component replacement algorithm in the traditional direct replacement of panchromatic image after matching,we use the NSCT decomposition of the panchromatic image,the multi-scale group,it can from the multi-scale and multi direction to a more accurate description of the panchromatic image features,further more,we use matrix low rank decomposition to decompose the multi scale group,this treatment can reduce the redundant information of the panchromatic image,the characteristics of the panchromatic image is more accurate;in the low frequency part,because it contains the spectral information of multi spectral images,we use spectral correlation coefficient to express the band ratio between,then the injection model will be characterized into.The experimental results show that the method proposed in this paper has a considerable effect on visual effect and quality index.(3)A new pan-sharpening enhanced spatial information enhancement algorithm based on deep learning is proposed.In this paper,the multi-layer auto-encoder network is used to extract the high-level features of the low-resolution panchromatic image and the high-resolution panchromatic image.The mixed residual is constructed in the hidden layer,and then the mixed residual is injected into the hidden of the multi-spectral image,to achieve the purpose of enhanced spatial information.Compared with the traditional pan-sharpening algorithm,the experimental results show that this algorithm has a better improvement in vision,and the details of its integration are more abundant.