Research Based On Pixel-Level Image Fusion Algorithm

Image fusion is to synthesize multiple source images with different information representation to one image, which makes the fused image to be more convenient for farther analysis. The essential of image fusion is a process that takes the advantages of redundant and complementary information of source images to make the fused image have more reliability, less ambiguity and more convenient for image analysis and understanding. As a novel technology in the filed of image processing and comprehension, image fusion has been widely applied to military affairs reconnaissance, medical diagnoses, remote sensing image processing, infrared target detecting and robot vision, etc.On the basis of reviewing the traditional image fusion algorithms, multi-directional and multi-resolution transform used in image fusion, such as Wavelet, Curvelet, Contourlet are studied detailedly, and Nonsubsampled Contourlet is introduced in image fusion firstly. Meanwhile, a novel image fusion algorithm based on Multi-Directional Undecimated Wavelet Transform (MDUWT) and local multi-characters matching estimate is proposed.The main contents and contributions of this thesis are summarized as follows:(1)The spectrum warping and artificial edges in fused image based on wavelet transform algorithms are discussed according to the theory analysis of wavelet and the experimental results of image fusion.(2)The basic theory of multi-directional and multi-resolution image transform and its application in image fusion are introduced. Multi-directional and multi-resolution image transform could cover the limitation of the finite direction of wavelet in its high pass-bands detection for image fusion, and could effectively detect the multi-directional information in the high pass-bands, which is beneficial to image fusion. The basic principle of the nonsubsampled contourlet transform is researched, and introduced in image fusion firstly.(3)A novel image transform, MDUWT, with shift invariant is presented and is applied in image fusion. The transform is based on the idea of multi-resolution, and fully takes advantage of the redundant and shift invariant of undecimated wavelet transform and the advantage of the nonsubsampled directional filter bank in detecting multi-directional information of image, so based on multi-resolution, the transform could detect image's multi-directional high pass-bands.(4)A novel image fusion based on local multi-characters matching estimate is presented. The fusion rule, estimating local area's multi-characters, is beneficial to use multi-characters of local image to improve fused image.(5)A simple image fusion experiment demonstration system is developed based on Matlab6.5 and Visual C++6.0. The system is on the thought of hybrid programming by making use of the interface technique of Matlab6.5 and Visual C++6.0 to realize calling matlab function by Visual C++6.0, and implement image fusion algorithm.