Multi-spectral Image Fusion Method Based On EMD

Image fusion, which is an important and useful technique for image analysis and computer vision in recent years, is a technique to combine information from multiple images of the same scene, and the integrated image reflects information from original images, using it to analyze and judge the objects will be more accurately and perfectly.Hilbert-Huang Transform is first introduced by Norden E. Huang from NASA in 1998, it is a new theory for signal analyzing, it use Empirical Mode Decomposition(EMD) to decompose the signal into many Intrinsic Mode Function. Hilbert-Huang Transform is a adaptive transform based on local characteristics of the signal, so it is useful for the analysis of non-linear and non-stationary signals. In recent years, it is widely used in many fields, such as signal denoising, engineering fault diagnosis and earthquake physics and so on. Since the Empirical Mode Decomposition method has good effects in one-dimensional signal processing, scholars extend it to two-dimension. They introduce Bidimensional EMD method to two-dimensional signal processing and get good results.The prime task of the paper is to do image fusion based on EMD. Firstly, I use EMD to decompose the image into some Intrinsic Mode Function's and a residual. In the sift of Bidimensional EMD, four main problems are existed: search of the local maxima and minima, interpolation methods'option, the stop criteria of sifting process and the management of points at the borders. Taking images'characters and decomposition speed into account to solve the four problems properly: extract the extrema points by comparing the candidate data point with its nearest 8-connected neighbours, create envelope by spline interpolation based on Delaunay triangulation, the number of iterations of sifting process equals to 3, use mirror reflection to solve the border points. And then use some fusion methods to the Intrinsic Mode Function's. Explicit fusion principle is put forward in Chapter 5, for different frequency segment use different fusion principle, and it takes the character of images into account sufficiently.At the last part of the paper, using traditional Principal Component Analysis, wavelet method and EMD to do image fusion, and compare the final fusion image and some indexes of image fusion. From the fusion image, we can be find out that the new method strengthens spectral information and details, from some quantitive fusion indexes, we also can see more superiority over the other two methods, so the fusion scheme is effective.