Research On Shearlet Transform Based Image Fusion

In the course of nearly three decades, a wealth of research achievements has been obtained in the field of image fusion and they have been widely applicated in monitoring, medical and remote sensing areas, etc. Conventional image fusion methods are usually based on certain kind of multi-scale transform, such as various types of pyramid transforms, wavelet transforms, beyond wavelet transform and so on. Shearlet transform is a type of beyond-wavelet transform and has been raised and matured only in recent years. It has overcome the shortcoming of conventional wavelet transform that lacks the ability of direction representation. While at the same time, studies on Shearlet transform based image fusion are still quite limited at present. The current implementations of Shearlet transform are normally divided into frequency domain based Shearlet transform and spatial domain based Shearlet transform. This paper studies various image fusion methods based on different types of Shearlet transform and the research content is divided into the following four parts:General image fusion(GIF) frameworks based on Shearlet transform: Both GIF methods based on frequency domain based Shearlet transform and spatial domain based Shearlet transform are proposed. Compacly supported Shearlet transform(CSST) is a type of spatial domain transform. And the conventional implementation of CSST is shift-variant which leads to the artifacts in the fused images. This paper studies fusion method based on cycle spinning, which eliminates artifacts in fused images through post-processing means, as well as the fusion method that directly eliminates the nature of shift-variant property. The latter is utilizing based on dual tree compacly supported Shearlet transform(DT CSST) which is almost shift invarient, and then applying it to the image fusion method in order to enhance the quality of fused image. And several groups of data are used to improve the performance of the above mentioned methods.Multi-focuses image fusion methods based on dual tree compacly supported Shearlet transform: The paper introduces the multi-focus image fusion method based on DT CSST and directional decision maps. The method further improve the quality of fused image compared with the GIF methods.Remote sensing image fusion methods based on Shearlet with PCA or IHS transform: Remote sensing image fusion is also known as pan-sharpening. This paperintroduces remote sensing image fusion methods based on Shearlet with PCA or IHS transform and then evaluates the performance through IKONOS and Quick Bird satellites’ remote sensing images. Experimental results indicates that the performance of image fusion methods based on DT CSST is superior to those of the methods based on frequency domain based Shearlet, various wavelet or Curvelet transforms.Three-dimensional magnetic resonance image(MRI) fusion methods based on Shearlet transform: Many medical devices can capture and form the 3D images about the structure of human body. However, the conventional researches on image fusion usually focus on two-dimensional images and if they are directly applied to the 3D images, the information of the third dimension will be lost. In the paper, DT CSST is firstly extended into three-dimensions. Then three-dimensional medical image fusion methods based on tensor product are introduced, taking the type of organs’ inner structure into consideration. And this method belongs to the feature level method. Finally, the methods are assessed by using multiple groups of MRI T2 * and quantitative susceptibility mapping(QSM) of the human brain. The result is that method based on three dimensional DT CSST delivers superior subjective and objective performances compared with frequency domain based Shearlet transforms and various types of wavelet transforms.Through studies in the four areas above, it indicates that Shearlet transforms have extraordinary performances. Image fusion methods based on Shearlet transforms, no matter it is general image fusion methods or methods for special purposes, presents better performances compared with the conventional methods.