Deep Features-Based CRF Graph Matching For Tracking Of Densely Packed Cells

The shoot apical meristems(SAMs)also referred to as the stem-cell niche,is the most important part of the plant body.The SAM cells are imaged by Confocal Laser Scanning Microscopy and stored in image stack time series.Developing a computational platform capable of robustly tracking SAM cells is very critical to understanding the cell growth dynamics.The cells in the SAM are tightly clustered in space and have very similar shapes and intensity distributions,and there is much noise in the deeper layer image slices because of the absorbtion of laser energy,and the images can be translated,rotated and scaled in the imaging process,thus how to segment and track all cells in image stack time series can be very challenging.In this paper,we focus on how to improve the accuracy of cell tracking,and the main works are summarized as follows:First,in the cell imaging process,to keep the plant cell alive for a long period of time,it is necessary to limit its exposure to the laser.This results in low contrast of cellular image quality with noise,which poses significant challenges to the efficient and robust cell tracking in such cellular images.This paper proposes to use the multi-scale transform sparse representation image fusion technology to fuse multiple registered cell images into a single image.The algorithm transforms the image from multiple scales to obtain high and low bands,then fuses the high and low bands separately,and finally reconstructs the fusion results of the high and low bands to obtain the fusion result.Experiments on three image sequences demonstrate that image fusion is beneficial to improve image quality and improve final cell tracking accuracy.Second,for cell image segmentation,firstly,the cell image is denoised by the mixed filter,and then the image is segmented by the watershed algorithm.Experiments show that compared with other segmentation algorithms,the watershed segmentation algorithm can effectively segment most of the cell boundaries and obtain better cell boundary closure.Finally,in this paper,a conditional random field model based on deep features is proposed to track cells.Based on the Siamese network,this paper establishes an 9-layer convolutional neural network,which uses a large number of cell images to train the network.Then,the output of the last layer of the network is extracted as the deep feature of the cell,and the deep feature distance of the pair of cells is used as unary potential function of the conditional random field model.The local feature distance which is invariant when the cell image is translated,rotated and scaled is used as the binary potential function of the conditional random field model.The experimental results show that the algorithm proposed in this paper can achieve parallel and stable tracking of plant cells whether for registered image sequences or unregistered image sequences.