Extending the SIOX algorithm: alternative clustering methods, sub-pixel accurate object extraction from still images, and generic video segmentation

Abstract

This paper presents a practical approach for object extraction from still images and video sequences that is both: simple to use and easy to implement. Many image segmentation projects focus on special cases or try to use complicated heuristics and classificators to cope with every special case. The presented approach focuses on typical pictures and videos taken from everyday life working under the assumption that the foreground objects are sufficiently perceptual different from the background. The approach incorporates experiences and user feedback from several projects that have integrated the algorithm already. The segmentation works in realtime for video and is noise robust and provides subpixel accuracy for still images.