Study On Image Fusion Technology Based On Multi-resolution Analysis

Image fusion is one of the important parts of information fusion. A new image can be syntheticed with two or more images through specific algorithm in order to obtain more complete description of information about the target or the scene. Multi-resolution analysis plays an important role in the process of signal and image processing. Image fusion methods based on multi-resolution are extremly important algorithms, which is widely used in the pixel-level image fusion techniques at present. Because the process of fusion is carried out at different scales, spatial resolution and decomposition levels, fusion results with effect obviously improved can be gained.Image fusion algorithms based on multi-resolution analysis is deeply studied in this paper. The main work is as follows:(1) Chapter 1 elaborates the research backgrounds, significance and objective at home and abroad, and introduces the principle and general framework based on multi-resolution analysis, then studies the models of multi-resolution analysis and fusion rules in the framework. Sevel commonly used performance evaluation indicators are summarized.(2) The theory of median pyramid is studied including its concepts and implementation algorithms. A multi-focus image fusion algorithm based on median pyramid is proposed. Taking advantage of non-linear characteristics of median pyramid, information of details can be retained effectively. Combining the local neighborhood average gradient activity measure, the method gets better effect than laplacian pyramid, contrast pyramid and gradient pyramid, and increases the entropy of fusion image.(3) The theory of multi-resolution watershed segmentation is particularly studied including its concepts and implementation algorithms. The fusion rule based on segmented regions is particularly researched and discussed. A remote-sensing image fusion algorithm based on region segmentation and wavelet transform is proposed. Considering the region information of each source image, the method can increase the robustness, and the contour of the interesting objectives can be made clearer. The experimental results show that the present fusion scheme can achieve better fusion performance than image fusion algorithms based on IHS transform and PCA transform.(4) The theory of Contourlet transform and NSCT is studied, and an image fusion algorithm based on NSCT is proposed. The fusion rule based on edge is used for fusion of low-frequency coefficients, and the rule based on the relevance of local coefficients is used for high-frequency sub-band coefficients. The algorithm can capture the directional information of details and the geometric characteristics of images at different scales. The experiments have proved the effectiveness both in the fusion of infrared, visible images and multi-focus images.