Automatic scale and image selection for panoramic images
Erkut Arıcan, Tarkan Aydın, Kemal Egemen Özden · 2016
We are presenting our work on automatic zoom level detection and minimal representative input frames subset selection for panoramic image applications. Thanks to the proposed techniques, the user can record videos from a single vantage point with free rotation and zooming motions, without a prohibitive recording time limit. The zoom level (or scale) of the input video frames are computed by mean-shift algorithm in a hierarchical structure. In order to select a minimal representative subset for excessive frames on a given zoom level, a greedy iterative algorithm is devised. Image matching is achieved by local image features and projective transformations. In addition to our recently proposed “Multi-scale Panoramic Augmented Reality” system, we expect the approach to be useful for other panoramic applications.