#66: Moving object segmentation using the flux tensor for biological video microscopy

K. Palaniappan, I. Ersoy, and S. K. Nath

Lecture Notes in Computer Science (PCM), Volume 4810, pgs. 483-493, 2007

motion, fusion, tracking, active contours, features

PlainText, Bibtex, URL, DOI, Google Scholar


Time lapse video microscopy routinely produces terabyte sized biological image sequence collections, especially in high throughput environments, for unraveling cellular mechanisms, screening biomarkers, drug discovery, image-based bioinformatics, etc. Quantitative movement analysis of tissues, cells, organelles or molecules is one of the fundamental signals of biological importance. The accurate detection and segmentation of moving biological objects that are similar but non-homogeneous is the focus of this paper. The problem domain shares similarities with multimedia video analytics. The grayscale structure tensor fails to disambiguate between stationary and moving features without computing dense velocity fields (i.e. optical flow). In this paper we propose a novel motion detection algorithm based on the flux tensor combined with multi-feature level set-based segmentation, using an efficient additive operator splitting (AOS) numerical implementation, that robustly handles deformable motion of non-homogeneous objects. The flux tensor level set framework effectively handles biological video segmentation in the presence of complex biological processes, background noise and clutter.