Robust Plant Cell Tracking In Image Sequences Using Conditional Random Field

The research on plant growth based on optical microscopic imaging technology is of huge significance to the basic research of life sciences.Therefore,the research of cell tracking algorithm in confocal microscopical image data has very important theoretical and application value.The main research object in this paper is the cell image of the shoot apical meristems which collected by confocal microscopy.And the shoot apical meristem is one of the most important plant organs,all other tissues and organs of the plant are derived from the proliferation and differentiation of its cells,hence the study of SAM cells is conducive to humans to further grasp the growth of crops.This paper focuses on the study of plant cell image segmentation,registration,and plant cell tracking.The main difficulties lie in:(1)The cells are highly densely arranged in honeycombs,and there are too much noises in the images;(2)The shape and size of the cells are random,and the color and grayscale information are also very similar.Therefore,it is difficult to extract features with a di stinct degree of discrimination;(3)In the matching process of the early-stage algorithm,error accumulation is easy to occur.(4)In the process of cell division,there are jump changes in the characteristics of the cells,causing some difficulties in tracking.Therefore,in response to these difficulties,the main tasks and contributions completed in this paper are as follows: firstly,the watershed segmentation algorithm is used for image segmentation.The experiments proved the watershed segmentation algorithm can effectively segment most of the cell boundaries.And then,this paper proposes a feature point extraction method based on local topology structure,and uses registration method based on feature point matching to register.It avoids the problem of the adjacent cell labels corresponding errors which caused by image rotation,and it improves the accuracy of the found feature points,such an effective image registration work lays a solid foundation for the following cell tracking work.Finally,in terms of cell tracking,this paper proposes a conditional random field(CRF)dynamic matching method to track plant cells in noisy images by exploiting the tight spatial topology of neighboring cells in a multicellular field as contextual information.(1)A conditional random field model is established in the cellular image o f the honeycomb structure,and the cells are traced using the spatial relationship of the cells in close topological relationship with neighboring cells and the interrelationships between cells in adjacent frames in the time domain as context information.(2)At the same time,this paper proposes a dynamic adjustment matching algorithm by which the neighboring cells of the matched cells will be selected as candidate cells at each match,and then only the best one pair will be matched at one time.The method effectively improved the previous method in the process of matching error accumulation problem and improves the tracking accuracy.The algorithm makes full use of the cell's own local features,spatial structure information between neighboring cells,and constraints of two frames of images in the time domain.It can achieve robust tracking of cells in noisy image sequences.Compared with the previous method,the experimental results show that the proposed method can improve the tracking accuracy rate by 10% in noisy image sequences.