Compressive tracking moving cells in time-lapse image sequences

Abstract

Tracking the motion of cells in time-lapse image sequences plays a pivotal role in both research settings and clinical practices. In spite of their prevalence, automated cell tracking approaches are still facing several major challenges, including the effectiveness of cell detection, accuracy of tracking and high computational complexity. In this paper, we propose a segmentation-based compressive tracking (SBCT) algorithm for moving cells. This algorithm consists three major steps, including detecting the bounding box of each cell, extracting image features in each bounding box using compressive sensing, and identifying the correspondence between cells in adjacent frames using a trained naive Bayes classifier. The proposed SBCT algorithm has been evaluated against seven state-of-the-art cell tracking approaches on two time-lapse images sequences provided by the 2014 cell tracking challenge. Our results suggest that the proposed algorithm can successfully tracking moving cells with relatively high accuracy and low computational complexity.