Near-Optimal Mosaic Selection for Rotating and Zooming Video Cameras

Applying graph-theoretic concepts to solve computer vision problems makes it not only trivial to analyze the complexity of the problem at hand, but also existing algorithms from the graph-theory literature can be used to find a solution. We consider the challenging tasks of frame selection for use in mosaicing, and feature selection from Computer Vision, and Machine Learning, respectively, and demonstrate that we can map these problems into the existing graph theory problem of finding the maximum independent set. For frame selection, we represent the temporal and spatial connectivity of the images in a video sequence by a graph, and demonstrate that the optimal subset of images to be used in mosaicing can be determined by finding the maximum independent set of the graph. This process of determining the maximum independent set, not only reduces the overhead of using all the images, which may not be significantly contributing in building the mosaic, but also implicitly solves the “camera loop-back” problem. For feature selection, we conclude that we can apply a similar mapping to the maximum independent set problem to obtain a solution. Finally, to demonstrate the efficacy of our frame selection method, we build a system for mosaicing, which uses our method of frame selection.