Research On Estimation Of Local Geometric Transformation In Image Stitching

Image stitching is the synthesis of the same scene with two or two or more narrow view images in the same scene. It is widely used in image editing, panoramic maps creating, virtual reality and so on. Even the core theories and methods which the technology involved(such as image aligning and image fusion, etc) is also important in the field of computer vision and image processing.The classic method requires the input image is taken from the same viewpoint or the scene where we shoot are approximate plane scene, we can see it is very stringent. However, For the problem of parallax, after we stitch the images by using the classic method, we may get a image which has a series of problems for ghosting and fuzzy, and we may couldn't stitch the images at all.We study it depend on computing the local geometric transformation model, the contents and results of our research are as follows:Firstly. We proposed a method named local geometric transformation model estimation where it has little error in registration. In our paper, we choose the best model from a plurality of candidate model strategy, making the image registration errors at different locations at the same time to a minimum based on the hash value similarity measure. Experimental results show that our method is more better than theirs.Second. We study and realize the image fusion method common in the literature, including the weighted average fusion algorithm, Laplacian pyramid fusion algorithm and wavelet transform fusion algorithm. Then we compare their results and select one of them for the development of image fusion splicing system.Thirdly. We develop a of a image stitching prototype system which is based on the local transformation model. The system is combine C++ with OpenCV developed in VS2012 platform. The system realized from the input image, image registration to the full functionality of image fusion, experimental results show that the system has a certain practicality.