Multispectral Image Fusion And Its Evaluation Method

For different types of multi-spectral images with different characteristics of data and for different levels of image fusion, analyzing and summarizing the results of previous studies, based on the foundation of which multi-spectral image fusion algorithms and the evaluation methods to the quality of fusion results are researched in this paper. The mainly work and achievements are as following:(1) The basic principle of probabilistic neural network is studied, based on which a multi-input and single-output PNN network topology which is fit multi-source image fusion is designed, the output formula of the image fusion model is derived using the interpolation function of PNN with the principle of minimum error. Multi-spectral image fusion at pixel level is realized using the image fusion model.(2) D-S evidence theory, including the combination rules of conflict evidence and the way to access the mass function, and its application in information fusion and image fusion are researched. By choosing appropriate samples to build feature library, using Gaussian function to construct the basic probability assign function, the mathematical model of the evidenc theory is established, ultimately D-S evidence theory is successfully applied to the multi-spectral image fusion and classification.(3) Methods of evaluating image fusion are researched Based on the existing objective and subjective evaluation methods, for the characteristic of multi-spectral image fusion, image quality evaluation methods which are based on edge information and are without depending on the reference image are proposed. And these objective evaluation parameters are used to assess the fusion images appeared in this article, and the results show the advantages and disadvantage of the fusion methods.(4) The corresponding algorithms are designed to simulate the multi-spectral image fusion and evaluation methods mentioned above, whose performance analysis is carried on, and experimental results verified the feasibility and availability of the methods mentioned above in this paper.