Research On Key Technologies Of Video Image Fusion And The Implement Of DSP

With the wide application of sensors with different functions and characteristics,the amount of information acquired by the system has shown an explosive growth,and the diversity is presented.It is necessary to develop new information processing technologies to solve the problems faced by multi-sensors in combination.The concept was developed here.Image fusion is a branch of information fusion in the image field.In this paper,the key technologies of image fusion and real-time realization of video image fusion are studied.Multimodal image registration and pixel-level image fusion are included in the research on key technologies of image fusion.Real-time realization of video image fusion is achieved in DSP platform.Multimodal image registration is studied from the mutual information based and the future based in this paper.Image registration based on mutual information is through the global search for the optimal image transformation model so as to realize the high precision image registration.The global search for the optimal image transformation is related to the algorithms for finding optimal solution.The optimal solution of the related algorithms,included genetic algorithm and particle swarm optimization algorithm,are studied in this paper.By combining of the advantages and disadvantages above two algorithms in finding the optimal solution,a new hybrid algorithm is proposed.The image registration results of above two algorithms and the traditional serial hybrid algorithm of above two algorithms are compared,the results show that the proposed hybrid algorithm in convergence speed and the optimal solution on search ability have good performance.For feature based multimodal image registration,the symmetrical SIFT future based multimodal image registration algorithm is studied in this paper.The problems of two kinds of statistical strategies of orientation histogram,based on gradient amplitude and based on the gradient occurrence are discussed in this paper.The correct matching rate of multimodal image feature matching is improved by combining image matching result by above statistical strategies of orientation histogram.Finally,the process of calculating projection matrix based on feature point is introduced in detail.In this paper,based on the literature [56],which combines NSCT and RPCA in image fusion algorithm,DWT + PCA image fusion algorithm is proposed.And the NSCT + RPCA,NSCT + PCA,DWT + RPCA,DWT + PCA image fusion algorithms are compared on the subjective evaluation and objective fusion quality evaluation in detail.What's more,the convergence rate of four kinds of image fusion algorithms were compared.DWT+PCA has a good performance in image fusion and real-time capability,and it is easy to achieve the real-time implementation of fusion algorithm on DSP platform.Finally,the real-time realization of video image fusion is studied.The knowledge of the hardware system and the DSP platform TMS320C6678 are introduced in this paper.Then the optimization based on DSP platform is introduced and discussed in detail.Through code optimization and multi-core parallel way,the realization real-time video image fusion is achieved.