Research For Method Of Image Synthesis In Multi-angle Scence

As an emerging technology involved many other fields, image mosaictechnology aimed at completing multiple images which have overlapping area togenerate a natural composite. Image mosaic technology widely used in aerospaceimages, remote sensing image stitching, virtual scene reconstruction, medical imageprocessing. In recent years, with the popularity of wireless video camera equipment,a large number of low-quality image data constraints the development of imagestitching techniques. Due to automated imaging devices which often have exposuredifferences lead to ghosting and obvious stitching boundary, in this paper, weresolve these problems from image registration and image fusion both aspects.Image stitching rate restricts its application in real-time systems, we also introducesthe issues which can improve stitching speed. In the past, stitching algorithmsperformance depend on subjective evaluation criteria which have arbitrariness andsubjectivity, we proposes objective evaluation criteria to contrast stitching images.Detail description of a few study questions as following:Exposure difference between the original images often leads to mismatch. Alarge number of mismatch points would lead to incorrect image transformationmodel which could cause serious ghosting problem. We use the global registrationalgorithm---bundle adjustment algorithm which selects reference plane and bundlesoptimization. In previous studies, the select of reference plane based on the numberof overlapping images, the area of the overlap region and the number ofmatched feature points within the overlapping area. We combine overlapping areaand the number of feature points as density of feature points within overlapping areaso that we can avoid setting the weight between area and the number of featurepoints and eliminate local mismatch impact in order to improve the stitching imagequality.In image fusion stage, we combine block-based exposure correction algorithmwhich resolve fusion ghosting caused by exposure difference and multi-band blendalgorithm which smooth the image boundary. Exposure correction algorithm basedon the importance of mapping weights superimposed image blocks in overlap regionto resolve structure incoherence more flexible. Multi-band blend algorithm resolvescolor incoherence.In this paper, we calculate the overlap region between the mosaic image and theoriginal images by calculating polygon intersection method. Apply color coherencecriteria and structural coherence criteria to evaluate the performance of the proposedalgorithm.