Font Size: a A A

Research On Pixel-level Image Fusion Method

Posted on:2020-07-19Degree:MasterType:Thesis
Country:ChinaCandidate:Y ZhangFull Text:PDF
GTID:2428330575981222Subject:Computer technology
Abstract/Summary:PDF Full Text Request
Image fusion technology refers to the method of merging information from multiple original images into the fusion image by means of various information processing methods.The fusion image combines the characteristics of each data source,and the target characteristics are more comprehensive,and the information utilization rate of the original image is higher.Image fusion is divided into three levels: pixel level,feature level and decision level.The focus of this paper is pixel level image fusion.Aiming at the two common applications of structure and function image fusion in medical research and panchromatic and multispectral image fusion in remote sensing research,a set of fusion algorithms are proposed respectively.For the fusion of structure image and function image in medical research,an adaptive decomposition algorithm based on edge perception filter is designed to separate the structure image into smooth layer and texture layer by constructing ideal target image.In order to avoid the instability of the decomposition layer and improve the decomposition efficiency,the decomposition accuracy index is constructed,and it is used to measure the separation degree of smooth layer and texture layer.By setting threshold,after each decomposition,the final decomposition times is determined.In the process of fusion,a dynamic fusion rule is proposed,which enables the fusion image to retain the original information of the structure image and the function image in the area where the structure image and the function image are separated.In the area where the structure image and the function image overlap,the spectral information of the smooth layer and the function image separated can be calculated dynamically,and finally the texture can be superimposed.Experiments show that the dynamic fusion algorithm based on edge perception filter is stable,and can not only retain the structure and function information of the original image,but also overcome some problems such as spectral distortion,region partitioning,and have a good subjective perception.For the fusion of remote sensing image,a multi-level pyramid fusion method based on gradient decomposition is proposed.According to the characteristics of remote sensing image,a decomposition filter core is designed according to gradient principle.Firstly,the filter core is used to decompose the panchromatic image,and the corresponding frequency subbands are separated iteratively.At the same time,the panchromatic image is sampled down.When the resolution of panchromatic image is equal to that of multispectral image,the panchromatic image and multispectral image are decomposed and sampled simultaneously,and the decomposition filter core is recalculated at each decomposition level.After decomposition,according to the characteristics of each level,the fusion rules are designed respectively.The energy information reflects more low-frequency levels,and the energy formula is constructed to fuse the corresponding low-frequency subbands and remote sensing images.At the intermediate-frequency level,the initial fusion evaluation is carried out according to the structural similarity index,and the second fusion is carried out according to the initial evaluation results,aiming at the high detail information,frequency levels are fused using overlay rules.Finally,the fusion image is obtained by inverse transformation.Experiments show that the multi-level pyramid fusion algorithm based on gradient decomposition retains the detail information of panchromatic image and the spectral information of multi-spectral image.It is stable for various types of remote sensing images and has better objective evaluation.Compared with other similar algorithms,it has certain advantages.
Keywords/Search Tags:Image Fusion, Medical Image Fusion, Remote Sensing Image Fusion
PDF Full Text Request
Related items