Abstract
A fast image stitching algorithm based on improved speeded up robust feature (SURF) is proposed to overcome the real-time performance and robustness of the original SURF based stitching algorithms. The machine learning method is adopted to build a binary classifier, which identify the key feature points extracted by SURF and remove the non-key feature points. In addition, the RELIEF-F algorithm is used for dimension reduction and simplification of the improved SURF descriptor to achieve image registration. The threshold-based weighted fusion algorithm is used to achieve seamless image stitching. Finally, several experiments are conducted to verify the real-time performance and robustness of the improved algorithm.