Multi-Cell Tracking Techniques Based On Ant System And Its LabVIEW Implementation

With the rapid development of the digital image processing and computer vision technique, the quantitative analysis of microscope cell image sequences has been widely studied, which can provide the references for medical research. In recent years, cell position and contour estimating analysis get more and more attention, and various techniques are developed as well, but there are still some of the reasons that restricts the development of multi-cell tracking technology, such as the low-SNR images, the uncertainty cell moving.Ant system, the representative of swarm intelligence has become a hotspot in the research of distributed artificial intelligence. The multi-ant colonies can cooperate and compete together to estimate multi targets’ parameters on the base of the strong searching ability of an ant colony. The deterministic tracking methods, represented by level set and active contour model, have been widely used in the field of tracking, due to their complete theories. In this thesis, we propose several multi-cell tracking approaches by combing the ant system with deterministic methods. The proposed approaches aim to accurately and jointly estimate the positions and its contour of each cell, and are built in Lab VIEW for the first time. The main contributions of this thesis include:1. For the difficulty to estimate the multiple cell parameters under complex scenarios, we propose a mixed multi-cell tracking approach by combining level set and ant system together, to make full use of their advantages. Birth ants are directly distributed into the regions depicted by raw curves achieved by the traditional level set evolution. The resulting pheromone field is embedded in level set to drive the evolution of cell curve to yield an accurate one and correspond cell position estimate. So, the searching efficiency and the tracking accuracy are improved. The experiment results show that our method could automatically track multi-cell and achieve an accurate contour estimation of each cell for given image sequences.2. For the difficulty to be convergent to the global optimal solution, we propose a mixed tracking approach based on the similarity between ant system and active contour model. Thus, this proposed algorithm provides a new approach to solve the active contour model, as well as accurately estimates the multiple parameters. Cell centroids are estimated by ant system, while the searching space is built on the base of the centroids. This algorithm integrates the characteristics of active contour model into ant system and well defined energy and heuristic functions. Consequently, the problem of cell contour estimation is actually converted to search for the marks of cell contours by group of ants. Experiment results show that our proposed approach is more effective than several existing methods.3. LabVIEW has some unique characters such as friendly interface, reliable performance, flexible operation, simple structure and so on, so we built multi-cell tracking system in LabVIEW by modular and graphical designing idea, to expand the applicability of our proposed approaches. The tracking system is valid and efficiency for given multi-cell images. This study will enable the approaches with a certain scientific value and business prospects.