Structure Tensor Based Image Fusion

Image fusion is a promising technique as an emerging discipline. The aim of image fusion is to get more comprehensive, more exact and more reliable image description of the same scene or object by integrating and distilling the source images from different sensors. Currently, image fusion has been widely used in remote sensing, medicine, military, computer vision, etc.First we introduce the background, significance, the basic processes and current problems of image fusion; summarizing the subjective and objective evaluation criterion; reviewing previous method in common uses, in which, image fusion based on wavelet transform is a popular method. On the basis of wavelet fusion algorithm we propose a new image fusion method by using structure tensor. The main contributions of this Master thesis are as follows:(1) On the basis of the existing wavelet fusion algorithm, through in-depth analysis and improvement, this paper presents an image fusion algorithm based on linear structure tensor, in which, the coefficients of the high-frequency sub-band is selected with the maximum of local coherence of structure tensor, and low-frequency sub-band using the weighted average.(2) A new algorithm is presented by using TV flow of NLST , which is an image fusion algorithm based on multi-scale TV flow of nonlinear structure tensor. In the scheme, coefficients of the high-frequency sub-band are obtained by weighted average.(3) Finally, on the basis of the image fusion algorithm based on multi-scale TV flow of nonlinear structure tensor, this paper presents an image fusion algorithm based on multi-scale the trace of nonlinear structure tensor, in which the coefficients of the high-frequency sub-band are obtained by a regional average.All algorithms have been carried on with Matlab. The experimental results show performance of these methods.