Research & Implementation Of Image Mosaic Technology Based On Fast Registration Using Feature Points

In recent years, image mosaic technology continues to develop. Image mosaic is the process of combining multiple photographic images with overlapping fields of view to produce a segmented panorama or high-resolution image. With the rapid development of computer vision, image mosaic technology has been greatly improved in accuracy and robustness, and is widely used in virtual reality, medical analysis, remote sensing satellites and aerospace. However, in emergency issues, such as firefighting, rescuing and public security, a more efficient kind of algorithm is needed. This paper focuses on the key aspects of image mosaic technology-image registration algorithm based on feature points, and improves the image registration efficiency under the premise of high accuracy and robustness.The main contribution of this paper is,1) Comparing the scale-invariant feature transform algorithm between SIFT and SURF and improving the efficiency of SIFT by optimizing the calculation process of the main direction, and reducing the dimensionality;2) Presenting a fast and high-quality image registration approach using BBF algorithm for rough match, and RANSAC algorithm, which is optimized by adding initial inspection and error eliminating method, for accurate match;3) Proposing a two-band blending method after balancing between linear blending and multi-band blending for better efficiency;4) Conducting contrastive experiments between our method and the traditional method, proving that our method can achieve stable, accurate and fast result by comparison;5) Implementing our method by building a system using unmanned aerial vehicle in order to create panoramic images from the air.