Research On Robust Tracking Of Plant Cells Based On Local Graph Matching

Plant cell research plays an important role in the fields of biology and medicine. Through the analysis of the cell image data such as cell segmentation and tracking, biomedical researchers are able to analyze the cell size distribution and motion characteristics of the cells by tracking the trajectories of cells.In this paper, we propose an automated segmentation and tracking system for the shoot apical meristem(SAM) cells. Watershed segmentation is a suitable method for SAM cells, but it has an over-segmentation problem. In order to remove such image noise, we have designed a mixed filter, which combined the advantages of frequency filter, spatial filter and morphological filter. The mixed filter contains wavelet denoising, average filter and erosion operator.The cells are properly segmented and then tracked by using a proposed Triangle Neighborhood Structure matching method, which exploits the cells'spatial context to find the seed cell pair for initial matching. The cell division tracking output acts as an indicator of the quality of segmentation and, in turn, the over-segmentation and under-segmentation errors are corrected by a local adjustment method, which is proved to be much more accurate and efficient than the global correction method.In the long-term tracking process for cell image sequence, a cell may disappear shortly because it is corrupted by noise or not correctly imaged, therefore, its lineage could be cut into several tracklets which should be associated together. We propose a method to associate such cell tracklets by searching the optimal predictions in the feature space based on Markov Chain Monte Carlo Metropolis-Hastings sampling methods.Finally, the experimental results on multiple datasets validate the effectiveness and robustness of our proposed cell tracking algorithm.