Research On Digital Image Fusion Algorithm Based On Multi-scale Transform

As a method of image processing, image fusion is to synthesize two or more image information into one image, which presents more complete information of the scene. The fusion rule will directly affect the visual effect of fusion image and the judgment of the scene,so the selection of image fusion method is very important. Image fusion has a wide application, such as target detection, tracking and recognition, scene perception, airport navigation, medical imaging and diagnostic, human visual aids, etc. The fusion rule of high frequency components that are obtained by multi-scale decomposition attracts so many scholars to study for a long time,but the low frequency component contains the most energy of source image, and it's the approximate image of source image, so the selection of low frequency component fusion rule is worthy to give full attention.Multi-sensor image fusion algorithm based on the analysis of the multi-scale decomposition technique is studied in this thesis. Including multi-scale decomposition tools such as pyramid transform, wavelet transform (WT) and Non-Subsampled Contourlet Transform (NSCT). Simulation experiment of multi-focus image fusion, infrared and visible light image fusion were performed, and the existing algorithm was improved. The main work and research is as follows:Firstly, the summary, classification and quality evaluation of image fusion are briefly introduced. Theoretical analysis and experimental simulation of pyramid transform, wavelet transform and NSCT were performed.Secondly, a new fusion rule based on the analysis of multi-focus image features was proposed. Considering the block centered characteristics of multi-focus image, and combined with WT,a new multi-focus image fusion methods was proposed based on WT and Block-dividing. The method mainly consider the fusion rule of low frequency component, and take full advantage of the characteristics of multi-focus image, using block operation instead of pixel operation. The reasonable block size of multi-focus image named Clock was determined based on a large number of simulation results.Finally, a new infrared and visible light image fusion method was presented considering the characteristics of infrared image. According to the characteristics of infrared image, using gray theory to the low frequency component of infrared image after NSCT for object extraction. To get the low frequency component of fused image, different fusion rule is applied to the target pixels and non-target pixels. And the fusion image which has some improved evaluation indexes and has a good visual effects will be obtain by further operation.The influence caused by the threshold parameter ? to fusion result was studied and the appropriate values of ? were obtained with which the fusion images named UN Camp and named Dune could have a ideal comprehensive assessment.