Automatic biological object segmentation and tracking in unconstrained microscopic video conditions

Wang, X 2018, Automatic biological object segmentation and tracking in unconstrained microscopic video conditions, Doctor of Philosophy (PhD), Engineering, RMIT University.


Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Wang.pdf Thesis application/pdf 51.39MB
Title Automatic biological object segmentation and tracking in unconstrained microscopic video conditions
Author(s) Wang, X
Year 2018
Abstract Cell and small biological organism tracking research is of fundamental importance for the analysis of dynamic behaviour for assisting the development of many biomedical image related applications. With the rapid development of digitised imaging systems, the immense collections of experimental (microscopic) videos make it nearly impossible to manually analyse the obtained data. Therefore, recent research has drawn attention to building automatic tracking systems to track the movement of cells and small biological organism models using videos taken by microscopes.

Although general object tracking (such as traffic cars and pedestrians) has been studied for decades, existing general object tracking systems cannot directly be applied to cell and small biological organism tracking, due to the differences in the imaging devices and conditions of the targets. This research therefore investigates the novel application of computer vision techniques to reliably, accurately and effectively track the movement of cells and small biological organisms automatically.

Due to difficulties in generating video segmentation ground-truth, there is a general lack of segmentation datasets with annotated ground-truth (particularly for biomedical images). This work proposes an efficient and scalable crowdsourced approach to generate video segmentation ground-truth and develops a tracking ground-truth generation system. To illustrate the proposed approach, an annotated zebrafish larvae video segmentation dataset and three tracking datasets have been generated and made freely available online.

Automatic cell tracking techniques require accurate cell image segmentation; however, current general object segmentation techniques are susceptible to errors due to the poor microscopic imaging conditions, which include low contrast typical of cell microscopic images. This work proposes a novel image pre-processing technique to enhance low greyscale image contrast for improved cell image segmentation accuracy. An adaptive, shifted bi-Gaussian mixture model is matched to the original cell image intensity histogram for greater differentiation between the cell foreground and image background, while maintaining the original intensity histogram shape.

Small biological organism videos taken by microscope imaging devices under realistic experimental conditions have more complex video backgrounds than cell videos. This work first investigates single zebrafish larvae tracking using dense SIFT flow and downsampling techniques. Many existing multiple small organism tracking systems require very strict video imaging conditions, which typically result in unreliable tracking results for realistic experimental conditions. Thus, this research further investigates the adaptation of advanced segmentation techniques to improve the performance of small organism segmentation under complex imaging conditions.

Finally, this work improves the multiple object association method based on the segmentation module for the proposed system, to address object misdetection and overlapping problems. This system is then evaluated on zebrafish videos, Artemia franciscana videos and Daphnia magna videos, under a wide variety of (complex)
video conditions, including shadowing, labels, and background artefacts (such as water bubbles of different sizes). The tracking accuracy of the proposed system outperforms three existing tracking systems.

Thus, the work in this thesis has contributions in automatic cell and biological organism tracking, where the investigation studied the region-based segmentation dataset construction generalised for biological organisms, intensity contrast enhancement for micrographs, segmentation improvement by removing imaging constraints and the final tracking accuracy enhancement.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Engineering
Subjects Computer Vision
Image Processing
Signal Processing
Keyword(s) Automatic multiple object tracking
Biological small organism tracking
Time-lapse microscopic videos
Ground-truth generation
Segmentation
Multiple target association
Gaussian mixture model
Versions
Version Filter Type
Access Statistics: 13 Abstract Views, 8 File Downloads  -  Detailed Statistics
Created: Wed, 05 Dec 2018, 11:15:00 EST by Keely Chapman
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us