Fusion Algorithms Of Multispectral And Panchromatic Images

Remote sensing data have different spatial resolution, spectral resolution and time-phasic resolution. Image fusion technology can combine their respective advantages in order to make up for information loss of single image, enhance image information analysis capabilities, and improve target classification accuracy and dynamic monitoring range. For the some outstanding problems of multispectral and panchromatic fusion algorithm such as the conflict between improving spatial resolution and maintaining spectral quality, this thesis makes analysis and research from the pixel level, feature level and decision-making level, and then puts forward the corresponding solution. The main research content and innovative results are as follows:1. In the aspect of image pre-processing, with regard to the difficulty of structure extract in registration algorithm, a novel image registration algorithm based on space projection in Mallat wavelet domain is proposed. As for low-frequency components, space projection character matching is used, and then the results of matching are applied in the high-frequency components to achieve coarse-fine matching, while as for the highest resolution component, normalized cross-correlation matching is used , finally the simulation results are given and analyzed. Experimental results show that according to space projection principle, the proposed algorithm convert two-dimensional data into one-dimensional 0-1 character string to matching comparison, which not only can reduce search space, but also greatly decline registration time. In addition, the hierarchical search strategy with layer by layer can reduce the matching error and improve registration accuracy. Compared with the common registration algorithms of cross-correlation and Hausdorff distance, the proposed algorithm is dominant both in the registration time and matching accuracy, which is ready for image fusion in the next phase.2. In the aspect of pixel level fusion, with regard to low resolution of fusion image, two multi-scale decomposition tools—àtrous wavelet transform and discrete Curvelet transform are firstly introduced into image fusion field, and then with the advantages of translation invariance inàtrous wavelet transform and multi-orientation decomposition in Curvelet transform, a novel image fusion algorithm based onàtrous- Curvelet transform is proposed, which is combined with IHS color space conversion. Different weighted fusion rules are adopted according to coefficient features of high and low frequent layer. The conditional weighted fusion rules that regional characteristics product as active measurement and correlation coefficient as matching degree are adopted in high frequent section, while the average weighted rules are used in low frequent section. Finally, the fusion image can be achieved by IHS inverse transform. Experimental results show that the proposed algorithm can effectively improve spatial resolution of fusion image on the basic of retaining spectral information, and then fusion effect of this algorithm is better than that of other multi-resolution analysis algorithms. In addition, space and spectral quality usually depends on threshold selection, so choosing proper threshold should be based on different applications and characters of original images.3. In the aspect of feature level fusion, the current fusion rules that the largest absolute value selection rule the weighted average rules have some disadvantages such as loss of fusion information and sensitivity to noise. In order to overcome these shortcomings and let fusion image contain much information of original images, a novel image fusion algorithm based on fuzzy reasoning in NSCT domain is proposed, which is combined with IHS color space conversion. Fusion rule setting of this algorithm is confirmed according to reginal features of NSCT transform coefficients, through the introduction of fuzzy reasoning principle to determine weighted value of the corresponding coefficients of original images. Experimental results show that the proposed algorithm both in suppressing spectral distortion and improving spatial quality is superior to the current fusion algorithms based on multi-resolution analysis, and then can overcome the shortcomings of poor anti-noise capability in traditional fusion rules. On the one hand, as a new multi-scale geometric transformation, NSCT has better direction selectivity and translation invariance, which can fully reflect image geometric information. On the other hand, adopting the weighted fuzzy reasoning fusion rules can effectively solve some uncertainties problems, and then these rules can be also extended to medical and other areas of image fusion.4. In the aspect of decision-making level fusion, with regard to high computational complexity of multi-resolution analysis algorithm, a novel image fusion algorithm based on dual-channel adaptive PCNN is proposed. Firstly, the multispectral image is converted from RGB space to lαβspace that is more in line with color transmission character. Next, the achromatic channel (l )image and panchromatic image are adaptively decomposed by simplifying tralditional PCNN model and defining image definition as the coupled joint coefficient, and then the largest entropy ignition time series are sent to decision factor to achieve the new achromatic channel image. Finally, fusion image is acquired by lαβinverse transform onαcomponent,βcomponent of original multispectral image and the new achromatic channel component (l ). Experimental results show that the proposed algorithm can not only solve the difficult problem on how to set traditional PCNN parameters automatically, but also with view to the correlation among pixels set and noise impact, fusion effect of this algorithm is better than that of other mutiresoluion fusion algorithms such as wavelet transform both on subjective and objective evaluation. Meanwhile, this algorithm can reduce the computational complexity.