Image Fusion Based On Un-decimated Morphological Wavelets And Guided Filter

Image fusion is a modernized high technology, which is combined with signal processing, image processing, artificial intelligence and sensor technology. This technique can extract the complementary information between multiple images, as well as ignore the redundant information, which is beneficial to detecting and tracking the target.Currently,most of the image fusion researches concentrate in the multi-scale image fusion, which have mainly experienced three stages: pyramid transform, wavelet transform and super wavelet transform. Wavelet transform is an analysis tool to linear signal, which can only using linear approximation to describe the nonlinear feature of images, leading to inaccurate description of images. While for super wavelet transform, the non-subsampled contourlet transform is applied extensively to extract better edge, but using this method increases the amount of computation hugely, posing a barrier to the real time processing. In order to describe the nonlinear characteristics of the image quickly and accurately, the un-decimated morphological wavelets(UMW) decomposition is adopted, which can take the nonlinear characteristics of images into consideration and make the high-pass coefficients hold definite physical meaning at the same time. Meanwhile the guided filter technology is applied to reduce the halo phenomenon, thus optimizing the image edges. In this thesis, the existing fusion methods were analyzed, so were the following researches based on the UMW method and guided filter method:1) An image fusion algorithm based on weight optimization of guided filter is come up with. The base layer image is obtained by decomposing the source image via UMW. Besides, the detail image, a different image between the source image and base layer image, retains more structure and detail information. The weights of edges are highlighted after the weight map of detail image being processed by guided filter, reducing haloing phenomena and making the edge of fused image more obvious.2) Another image fusion algorithm based on the optimization for low frequency coefficients of guiding filter is proposed. In this algorithm, by adopting the multi-direction un-decimated morphological wavelet the decomposing images in each direction can possess a nonlinear transform characteristic. After decomposition, the low frequency coefficient contains more structure and detail information. By conducting fast guided filter to fusion those coefficients, it can not only meet the real time requirement, but also optimize the edge details of fusion image.According to the experimental results from infrared, visible images, multi-focus images, medical images and remote sensing images and the analytic results, from the subjective perception and objective indicators respectively, this method can be employed to keep the edges information of the image more intact and reduce edge halo more significant, booting the visual effect.