Research On Multi-Cell Tracking Techniques For Low-SNR Image Sequences

With the rapid development of digital image processing technique, the automatic cell recognition and tracking methods for microscope video image sequence receive comprehensive study and application, and it is valuable for several areas such as drug development, disease diagnosis, disease treatment, etc. In this thesis, the automatic tracking of biological cells in time-lapse low-SNR microscopy image sequences was studied. According to the characteristics of cell image including luminance variance of local image, low contrast between background or foreground and high density of noise, this thesis mainly focus on the work of multi-object tracking algorithms based on random finite sets(RFS) and swarm intelligence(SI) to achieve better tracking results. In addition, a multi-Cell Tracking Integration Software System which includes several stochastic tracking methods is designed.The multi-object tracking method based on RFS is one of the most active topics in the field of tracking. In view of the lack of quantitative comparison of two RFS-based multi-cell tracking methods, i.e. Gaussian mixture probability hypothesis density filter(GM-PHD) and multi-Bernoulli filter respectively, a sample cell detection method for the GM-PHD tracking method and a likelihood function based on RGB color histogram for the multi-Bernoulli filter tracking method were designed. By testing on the same image sequence, the differences and similarities between the methods and quantitative comparison of the experiment results are analysis to provide evidence of the choice of the two RFS-based tracking method for low-SNR microscopy image sequences.SI is a computing technology inspired by the law of biological group behavior. In this thesis, a novel method based on particle swarm optimization algorithm(PSO) is developed for automatic tracking of biological cells’ location and contour. Two different PSO-based tracking models are built to give the initial positions of the existing cells and emerging cells, respectively. A PSO-based contour module is then proposed to determine the corresponding contour of each cell and finally achieve a precise position tracking by an iterative centroid updating process. Furthermore, by introduced a Gaussian model, the tracking performance is good for collision.Finally, an integrated software system for multi-cell tracking is constructed. It contains several kinds of current popular randomness multi-cell automatic tracking method, which can be divided into three categories: methods based on random finite set theory, methods based on particle filter and methods based on swarm intelligence. In addition, the system contains an artificial tracking module, which can work out the data of total number and positions of cells. The system can track multiple cells for some particular image sequences, and generate detailed reports data, and analysis the data to compare the characteristics and performance of these different methods.