| Image fusion fully uses the redundant and complementary information of varioussources images, so that the fused image has higher credibility and resolution and is moresuitable for human visual perception and computer subsequent processing. Denoising isto minimize and remove noise under the premise of preserving image features anddetails. Image fusion and denoising are important steps in image pre-processing, thetreatment effect is good or bad has a significant impact for the subsequent imageprocessing such as texture analysis, feature extraction, pattern recognition, therefore theresearch for image fusion and denoising has a very important significance.NonSubsampled Shearlet transform (NSST) is the latest proposed multi-scalegeometric analysis tool, which has excellent characteristics such as sensitivity with verygood direction, good local frequency characteristics. Moreover, it is a near-optimalmultidimensional function "sparse" means, which may well determine its application toimage fusion and denoising. The main work of this thesis is as follows:1. Introducing the development of image fusion and denoising, giving the principle ofNSST and implementation process.2. Describing the general principles of pulse coupled neural network (PCNN), andproposing an image fusion algorithm which combines NSST with PCNN. In thisalgorithm, for the high frequency sub-band coefficients transformed by NSST, animproved spatial frequency is used as PCNN input, and the improved Laplace energy isadopted as PCNN link strength. The experimental results show that the algorithmproposed in this thesis not only has better results in subjective visual, but also has someimprovement in objective criteria.3. A NSST image denoising algorithm based on normal inverse Gaussian model isproposed. Firstly, the algorithm builds a statistical model for the high frequencysub-band coefficients transformed by NSST with the normal inverse Gaussian model forthe priori model and estimate the model parameters of each sub-band. Then underBayesian maximum a posteriori probability estimation criteria, the algorithm derives thevariable threshold function expression corresponding to normal inverse Gaussianmodel. |
