The Image Fusion Technology Based On Multiscale Geometric Analysis

Image fusion is the technology that combines sensors, image processing, signal processing, displaying, computer technology and artificial intelligence, and many other modern high-technology. The technology combines multiple images on the same scene information, and finally obtains one image about the same scene images. Image fusion plays an important role in medicine, surveying, geographic information systems, industrial, intelligent robots, and military fields, and becomes an increasing concern of many researchers.In the transform domain, image fusion based on wavelet is used more often now. Wavelet transform has its significant advantages in one-dimensional signal processing, but it can't take full advantage of the unique geometrical features of the data itself in processing natural image, and the directional information that wavelet captures is limited, so it is't the best way for the high-dimensional function containing line or surface singularity. Multiscale geometric analysis (MGA) is the optimal high-dimensional functions and is developed in recent years, mainly contains Ridgelet transform, Curvelet transform,Bandelet transform,Contourlet transform and NSCT. In this thesis,MGA is introduced to the pixel level image fusion,with studying the image fusion methods on the base of transform domain and fusion methods, a new improved fusion algorithm based on NSCT and PCNN is proposed. The algorithm uses the spatial frequency and directional contrast to trigger PCNN. By comparing the firing times of different PCNN,we can get better coeffieients. The algorithm keeps the texture details effieiently and improves the fusion results a lot.Specifically, this paper describes the theroy of Contourlet transform,NSCT and WBCT in detail and their application in image fusion. The paper focuses on image fusion based on NSCT, and introduces the application of PCNN in image fusion and proposes a new fusion algorithm of combination of NSCT and PCNN which can effectively preserve the source image details and edges. Through the simulation of multi-focus image and infrared remote sensing image fusion, the algorithm is validated.Evaluation of the image fusion effect is also an important field. Finally, the evaluation of the subjective and objective method for image fusion is proposed. By comparing the simulation results of wavelet transform, Contourlet transform, NSCT and wavelet-Contourlet transform, indicates that NSCT has relatively superiority of other methods and the fusion image is more clearly. Then, the comparion of the results shows that the new improved fusion method greatly improves the edge, texture, and retains more source information, so it is a more versatile strong and effective fusion.