Fusion Method Based On Multi-resolution Analysis Of Multi-source Image

As one of the important aspects of the multi-sensor information fusion, the image fusion technique is how to use different images from different sensors, through some algorithm, to obtain a more accurate and abundant new image, then we can gain more exact, complete and reliable description and decision-making. The image fusion technique is used widely in military and civil application; it's a very popular research domian.The image fusion algorthim has three fusion levels: pixel-level, feature-level and decision- level. Pixel-level fusion which is the base of feature-level and decision-level is the elementary level, and is the main contents of this thesis. This thesis mainly discusses the multi-resolution image fusion algorithms in pixel-level. First, we introduce the background of research, development status, application in domestic and overseas, we also elaborate the hierarchy and class of image fusion. Second, we focus on the image fusion algorithm based on the multi-resolution analysis. We introduce the fundamental principles, the process of analyze and synthesize, fusion step and characteristic. We also summarize the main image fusion rules, objective and subjective evaluation criteria for image fusion performance. In addition, we advance an improved fusion rule of wavelet decomposition's coefficients based-on local variance/gradient's select and weighted average. Third, we study how the number of decomposition levels, different fusion rules and decomposition mode influence the image fusion performance, and get some important conclusion. And then, we analyse the principles and characteristic of lifting non-separable wavelet transform based on non-separable wavelet, and use it fuse different images. Simultaneity, because the lifting non-separable wavelet transform don't have the property of shift-invariance, we propose the redundant lifting non-separable wavelet transform, and use it fuse images, the results is perfect. At last, we summarize the work, and present deficiency and research orientation in feature.This thesis selects many fusion images of different type, makes plentiful emulate experiments, compares and analyzes the results of emulate experiments, and gets some conclusions which have important referenced value. There is two innovative points: one is advancing the improved image fusion rule: local variance/gradient's select and weighted average; the other is proposing the image fusion algorithms based on the redundant lifting non-separable wavelet transform. The examples indicate that they improve the performance of image fusion at some hand, and have some application value.