Research Of Image Fusion Based On Multi-objective Optimization Theory

The technology of image fusion is so important that it is used for many areas of human beings careering. In homes, the research of the technology is behind than foreign countries, it has still staying the level of theory researching. How to improve the quality of pre-treatment and fusion; how to enhance the efficiency of fusion; how to form the united fusion framework; and build more objective evaluation index, which are hotspots in the area of the image fusion.Firstly, the paper dose some research and simulations for traditional ways of image fusion, and accumulates some exercises and ways of that. Secondly, the paper does some research for pre-treatment of image fusion, especially the part of image denoising. There is an inspiration acquired by the theory of multi-objective optimization algorithm. The paper proposes a new kind of improved NEMOPSO algorithm. The new algorithm based on multi-objective particle swarm algorithm framework. However, there are some differences between them. The new algorithm adopts more effective ways of speed changing and multi-objective choice processing which makes better performance. The new algorithm has been used for image denoising and image fusion, which have achieved better results. Some main innovation points of whole work are shown as follows:First, it proposes a new algorithm named NEMOPSO based on the objective comparison. It jumped out of frameworks of linear weighting ways either adaptive variable weighting ways. It also not introduced the way based on adaptive grid or the way based on crowded degree and Pareto theory which are have high computational cost. Not directly polymerizing multi-objective function, not assigning priori weights due to personal preference, and not sorting processing or calculating crowded degree based on the concerned degree of multi-objective function values, but it is comparing the value of every index directly, it is only getting the comparison of the number of each objective which is larger or smaller than the others. This way has thought the comprehensive influence of each index, thus, it let the algorithm much close to the Pareto optimality. At the aspect of changing speed, the paper designs two times gravity changing functions, that makes differences among speeds of swarms and thus and avoid premature of swarms.Second, it proposed a new index (OR) for the evaluation of multi-objective optimization algorithm performance. The index is proposed based on statistics. Through calculating the velocity of particles which are over the maximum speed after the velocity changing each time, it gets the number, than divided by iterative particles and times which get the value of OR. This indicator reflects the ability of particle learning and global convergence. After many tests, I find that OR as well as target distance, separating degree have relevant increase or decrease trend, which means the new evaluation indicator is effective. The paper adopts the indicator to evaluate multi-objective algorithms boldly.Third, in the process of image fusion for multi-objective search range, paper proposes a new way which is not searching directly in the matrix of fused image, but in the component matrix concerned of decomposed image. Experiments show that this method not only greatly reduces the processing speed, and search result is also good.The paper introduces two random multi-objective testing functions which are calculating maximum and minimum values. There are relatively obvious contradictions among decision particles. Experiments proofs that the new algorithm can better convergence in Pareto front. Compared with classic multi-objective algorithm, the proposed algorithm really has better performance in separating degree, target distance, OR and running time. The paper also put a new algorithm for image denoising and image fusion, to compare with a variety of traditional algorithms, after lots of experiments and simulations, the thesis proposed method is proved is more excellent and equilibrium both at subjective and objective performance.