Research And Implementation Of Feature-based Image Mosaic

Image mosaic is a technology that fuses the mutual overlapping portions of imagesequences into a wide scene seamless integration of high-resolution images, and has beenwidely used in computer image processing, medical imaging and virtual reality research andother fields. In feature-based image mosaic, time-consuming is mainly on feature extractionand feature matching stage, while the choice of image registration and fusion method willinfluence the final mosaic effect directly.In this paper, we study the related technologies of image fusion splicing for imagescollected by camera rotate. By the comparison and study of current main methods, based onSIFT feature point, a method of optimizing and improving image fusion splice is proposed.During the matching phase, this paper removed external point (mismatching points) by addingconstraint conditions, which improved the probability of internal points and effectivelyreduced the random sample consensus algorithm iterations. The steady transformation matrixwas used for the image unified coordinate transformation. Finally, the weighted averagemethod was used to achieve image fusion mosaic. The experiment result shows that comparedwith traditional algorithms, the improved algorithm can greatly improve the probability ofinterior point which thereby improves the efficiency of image stitching and registrationaccuracy.As to the lack of improved algorithm on real-time matching, the algorithm is furtherimproved and optimized. A new algorithm is introduced which used a new feature extractioncalled FAST, to describe the feature points we use a binary descriptor called FREAK. Thenthe Nearest-Neighbor with Distance Ratio method is used to look for junior matching pointmeanwhile the external points are removed by the improved RANSAC algorithm. Finally, theweighted average method was used to achieve image fusion mosaic. The experiment resultshows that the improved algorithm can greatly improve the probability of interior point whichthereby improves the efficiency of image stitching and registration accuracy compared with traditional algorithms. And compared to the SIFT algorithm, the algorithm can greatlyimprove the efficiency of image stitching and could be used in real-time stitching.