Image Fusion Research Based On Visual Underlying Features And Multi-scale Analysis

The image information obtained by single image sensors is limited,so that it is hard to describe the sensors accurately.To solve this problem,the image fusion technology is developed.The technology could merge the useful information of single sensors image to get a high-quality image which contains more interesting information and has more advantage in post processing.In recently,various image fusion algorithms have been developed by many researchers and been widely applied in many field such as computer vision,robot technology,digital photography,automatic target recognition,military field et al.To make up for the defect of existing image fusion methods,this paper do a series of research about improving performance.The study results are as follows:For the infrared and visible image fusion,the paper proposed novel sharp measures for different subband coefficients respectively to overcome the defect that only single feature is considered in the fused rule in the NSCT domain.The paper considered the significance of a local structure,contrast information and brightness information in designing the fusion rules.For different subbands,different clarity measures are designed and the relative importance of different features are adjusted.For the multi-focused image fusion,to overcome the defects,the focused information can't be extracted accurately,exist in the traditional spatial domain-based fusion methods,this paper proposes a multi-focused image fusion method via multiscale images analysis technique and non-local means filtering.In which,a novel multi-scale images analysis technique is presented to integrate the advantage of different scale images.At the same time,we develop a new focus measure to identify the focused pixels by introducing the bilateral filtering idea.Finally,to mitigate the boundary seams produced by spatial domain methods,the block consistency validation and non-local means filtering is performed on the final decision maps to generate the fusion weight maps for source images.For the medical image fusion,to overcome the defects of consuming time and weak ability of expressing exist in the traditional KSVD-based learned dictionary,a novel dictionary learning method based on visual low-level features filtrating is proposed for the medical images fusion problem.According to the medical images features,the brightness information which reflects the energy features of images and the spatial frequency information which reflects the edge changing are regarded as two visual low-level features measures to filter image blocks to construct the training set.And then,the detail information dictionary and brightness information dictionary are trained by the KSVD algorithm respectively.Moreover,the medical images fusion frame is designed based on the trained dictionary.