Research On De-ghosting And Eliminate The Seam Of Image Fusion

Image fusion is a significant subject in the field of data fusion, its aim is to integrate the complementary information or significant information of the images (homologous or non-homologous) into one image which is integral and has complete information after extracting useful information and removing redundant information. On account of the sequential images change over time, or affected by registration and geometric transformation, the overlapping areas of the images exist diversities; There is ghosting and stitching seam if using the direct fusion method, therefore, how to eliminate the ghosting and the seam in the fusion process becomes two very important nodes. The best way to eliminate ghost and reduce the stitching seam is to find the best stitching line. After analyzing and testing those latest methods, we find some obvious limitations:1.The direction of the stitching line is not flexible, this will generate the ghost easily.2. The methods of searching stitching line only consider the horizontal shift and don’t apply to bidirectional shift or anomalistic overlapping area.3.The existing methods can’t avoid the stitching seam possibly during the process of de-ghosting.4.Those methods can’t avoid the generation of ghosting when eliminating the stitching seam.To those imperfections, the works of this article include two parts,1) finding the best stitching line,2) eliminating the stitching seam. The article proposed a method which not only could avoid ghost but also eliminate the stitching seam possibly. The method fully considers the relations include chromatic aberration and structure transformation between pixel and its neighborhood; It extends the search direction to the horizontal direction and could define the start-point and end-point, so this method could be used in irregular overlapping area and has a greater applicability. The wavelet image fusion method based on stitching line was proposed to eliminate the seam after splicing. In the fusion method of low frequency and high frequency, the fusion distance and fusion direction was added for avoiding generating ghosting again; Base on the stitching line, the method uses linear weighting method to gain the low frequency coefficient in the low-frequency domain, in the high-frequency domain, it gets the coefficient by the comparison of regional energy. The panorama was obtained after the inverse wavelet transform and embedded the fusion result into the overlapping area. The experiment section includes comparisons between our method and others, the result demonstrated that our method is better.