Research And Implementation Of Panoramic Image Mosaic Algorithm

Panoramic image mosaic stitches multiple overlapping images into a360-degree circular image which describes the scene information. Panoramic image mosaic is developing and gaining in popularity rapidly in recent years as a computer vision technology. It has a wide range of applications in space research, medicine, virtual-reality and normal life.This paper conducts in-depth study with panoramic image mosaics, and proposes an automatically image sorting algorithm based on the SIFT features and a SIFT feature matching algorithm based on the modified nearest neighbor algorithm. The concrete work is as follows:1. This thesis studies the history and state of the art of the panoramic image stitching at home and abroad, gains a depth insight of its key technologies, such as image feature extraction, image registration and image fusion.2. This thesis studies the SIFT feature extraction algorithm and proposes an automatically image sorting algorithm based on the SIFT features.3. This thesis studies image registration algorithm and proposes a modified nearest neighbor algorithm for SIFT feature matching algorithm. It calculates the cross-correlation coefficient of matching point pair and determines whether the matching point pair is matched according to the relationship of cross-correlation coefficient and default threshold T.4. This thesis studies the RANSAC algorithm to calculate projective transformation matrix and image fusion methods.This thesis uses the SIFT feature extraction algorithm to calculate the image SIFT features, and the modified nearest neighbor algorithm to match SIFT features and sort images. RANSAC algorithm is used to purify image SIFT features, and the least squares method is used to calculate projection transformation matrix, at last we fuse the images using the weighted average algorithm.