Research On Multi-source Image Fusion Algorithms

Multi-source image fusion is to comprehensive two or more images of the same scene that gained by multiple sensors, in order to obtain a new image which is richer in information, more accord with human visual characteristics or more suitable for computer processing. Its application has been throughout the airport navigation, security monitoring, medical imaging and diagnosis, machine vision, intelligent transportation, industrial control, military industry, etc. Therefore, the research on multi-source image fusion is valuable.The basic theory of image fusion, image fusion method based on multi-scale analysis and image fusion method based on compressed sensing theory are introduced in this paper. The main work of the full text is as follows.(1) The significance of wavelet transform and contourlet transform coefficients are analyzed, and the traditional image fusion rules based on multi-scale analysis are studied. The experimental results show that the contourlet have a better fusion result than wavelet in the aspect of keeping edge information.(2) For the feature that the focus area of multi-focus image may changes gently, traditional image fusion rules may enhance the SF to a wrong level. A improved fusion rule based on regional gradient consistency is proposed. The experimental results show that the fusion image have a higher PSNR.(3) For the characteristics of infrared and visible light image fusion. A enhanced fusion rule is proposed that low frequency based on regional energy and high frequency based on regional gradient, which can enhance the brightness of the fusion image.(4) Image fusion method based on compressed sensing theory are introduced. For the low frequency coefficient is not sparse which has bad effect to the image reconstruction, this paper proposes a fusion method based on compressed sensing of DWT high frequency coefficient. Experimental results show that the layer number of wavelet decomposition is reduced and the fusion effect is improved, so the amount of calculation is reduced.(5) For the problem that gaussian matrix is hard to realize for hardware, a kind of matrix which only contains minus ones, zeros and ones is designed and optimized. The experimental results show that the proposed method can obtain a better fusion result in the case of fifty percents sampled.Finally, the full text is summarized and prospected.