The Technology Research Of Image Mosaic Based On Improved SIFT Algorithm

Image mosaicing technology as an important branch of digital image processingcan well solve the contradictions and problems of vision and resolution, it was widelyused in the panoramic image synthesis,3-D scene reconstruction, medical imageanalysis, computer vision field and video retrieval, compression and other fields.Therefore, Researching on image stitching technology had a very important practicalsignificance.This paper studied the algorithm based on improved SIFT the image stitchingtechnology. Firstly, it introduced the significance of the technology, current situationand its applications, and then, summarized the image mosaic and the key technologies,describes the image acquisition, image preprocessing, image registration and imagefusion technology. This paper focused on image registration and image fusiontechnology, and proposed a depth research and analysis to the classic image registrationalgorithm-SIFT (Scale Invariant Feature Transform) feature points matching algorithm.In order to improve the slow speed of the SIFT algorithm, this paper presented asimplified SIFT algorithm, this method firstly used the original SIFT algorithm to detectfeature points, and then used the characteristics of the rotational invariance of thecircular window, ignored the front two steps to one step of the allocation direction of thefeature points and reduced the dimensionality of algorithm, meanwhile used thetwo-way matching to ensure that the algorithm accuracy,Then used three pairs ofmatching point to calculate the transformation parameter, at last adopt the image fusionmethod which based on wavelet to obtain the image stitching result.The improved SIFT algorithm introduced in this paper, not only has the advantagesof the original algorithm, such as: the right translation, rotation, zoom scale, noise, evenillumination changes has a good robustness, but also the speed of the improvedalgorithm was faster than the tranditional SIFT algorithm nearly one time. Theexperimental results showed that the algorithm can get satisfactory results in the imagerotation, translation, scaling and noises.