Key Techniques Research Of Image Mosaic Based On Feature Points

Image mosaic is an important research direction in image processing field. and has beenwidely used to such fields as computer vision, remote sensing technology, intelligenttransportation, etc.. However, image mosaic algorithms at present commonly have highcomputational complexity and poor mosaic effect. So this paper focus on the huge originaldata set, a lot of iterations and many false matches exist in the traditional RANSAC algorithmwhich be used in image registration, and the problem of a large number of ghost caused byusing traditional Fade in-Fade out algorithm to fuse the images which overlap in large degree,proposing corresponding improved algorithms and verify the effectiveness of the improvedalgorithm by experiments. The major work of present paper is as follows:（1）Constructing an image mosaic structure based on feature points. After pre-processing thereference-stitched images, detecting the feature points by SIFT algorithm, then achieving thepreliminary matching of features by the k-d tree and nearest neighbor search algorithm, andthen using an improved RANSAC algorithm based on the geometric constraint to registryimages, finally using the bilinear interpolation method and the Fade in-Fade out algorithmwith the correction of overlap region to achieve image fusion.（2）In-depth study and analysis of the SIFT feature points detection algorithm and Harriscorner points detection algorithm, comparatively analyze the two algorithms in such foraspects as the number of feature points and matching points, matching rate and running time.（3）Proposing an an improved algorithm based on the geometric constraint. Focusing on thehuge original data set, a lot of iterations and many false matches exist in the traditionalRANSAC algorithm, this paper proposes an improved RANSAC algorithm. Using two kindsof relationship between matching points: one is the Euclidean Distance between the couple ofmatches is equal and another is the slope of the line between each couple of matches is equal,forming a geometric constraint model to pre-pure the original data set in RANSAC algorithm,to reducing the number of the data set and the false matching therein, and decreasing theoperation times and calculation amount.（4）Putting forward an improved Fade in-Fade out algorithm by using a threshold to modifythe overlap region. Focusing on the problem of a large number of ghost caused by usingtraditional Fade in-Fade out algorithm to fuse the reference-stitched images which overlap inlarge degree, this paper proposed using the threshold to modify the overlap region, then using the Fade in-Fade out algorithm to complete image fusion, thereby eliminating the ghost andgetting the ideal fusion results.This paper calls the OpenCV image processing library based on the Microsoft VisualC++development environment, developing the whole image mosaic algorithm frameworkwith C language, verifying the algorithm proposed in the paper with a lot of experiments. Bycomparing the traditional RANSAC and the improved algorithm in the number of matchingpoint pairs, iterations, the multiplication and computation, verifying the improved algorithmcan efficiently improved the efficiency. According to comparing the mosaic result, verifyingthe proposed improved Fade in-Fade out algorithm by using a threshold to modify the overlapregion can greatly eliminate the ghost. The experimental results show that the proposedmosaic algorithm architecture can complete mosaic of different scenes, it costs shorter timeand can get desired mosaic result.