Full text on Karlsruhe Institute of Technology Library
X-ray imaging is a genuine tool to reveal internal structures of opaque objects. This is possible due to the penetration properties of its probe - X-ray radiation. Modern synchrotron facilities, equipped with high-resolution detector systems, provide X-ray radiation of unique quality and allow to investigate a broad range of dynamical processes, both in materials and biological specimens. To perform automated and quantitative analysis of time-resolved X-ray data, a method capable to retrieve dynamical information is required. In this work we develop a general-purpose framework for X-ray data analysis based on optical flow.
Optical flow methods traditionally belong to the field of Computer Vision. Finding correspondences between time-lapse images is a key problem in a variety of applications such as robot vision, tracking systems and video analysis. In the scope of this work we adapt variational optical flow methods - a specific class of approaches used to determine the optical flow - to the task of X-ray data analysis.
The quality of time-resolved X-ray data is diverse, ranging from high-resolution datasets to low-contrast, noisy images with artifacts. We provide a detailed classification of X-ray data. This taxonomy serves as a reference point for the development of image preprocessing, motion estimation and data analysis techniques. Image preprocessing is employed to enhance the original (raw) X-ray data in order to improve the accuracy of optical flow estimation for the case of challenging data.
To develop an accurate and robust motion estimation model, we perform a systematic evaluation of state-of-the-art optical flow techniques and make quantitative performance analysis of their components.
On the top of the optical flow estimation we provide an extensive data analysis toolkit including automated tracking, flow analysis, motion-based segmentation, image registration and detection of temporal changes. All the devised techniques can be applied in 4D (3D + time) to enable analysis of tomographic data. The implementation of the developed techniques incorporates advanced numerical schemes and computations on GPU. Thereby, the processing of a vast amount of X-ray data is feasible.
Finally, we present the application of the optical flow methods to a number of scientific problems from various research fields. These examples include flow analysis and particle segmentation in semi-solid alloys, analysis of morphogenesis in living frog embryos, coalescence events estimation and stability studies during the foaming process, and tracking of morphological dynamics in living insects.
Figure: Experimental setup for phase-contrast X-ray microtomography.
Figure: Cell motion analysis from 3D time-lapse series of Xenopus laevis embryo during mid-gastrulation.
Figure: Morphological dynamics and kinematics analysis of the screw joint during defensive movement.
- Introduction
- 1.1. Motivation
- 1.2 Outline
- 1.3 X-ray Imaging
- 1.4 Variational Optical Flow
- 1.5 Aims of the Work
- Optical Flow Methods
- 2.1 Modeling
- 2.2 Numerical Solution
- Data Analysis Framework
- 3.1 Data Preprocessing
- 3.2 Analysis Based on Optical Flow
- Computational Framework
- 4.1 Software Framework
- 4.2 High Performance Computing
- 4.3 Visualization
- Evaluation on Synthetic Data
- 5.1 Data Taxonomy
- 5.2 Experimental Data
- 5.3 Optimization of Parameters
- 5.4 Quantitative Evaluation
- 5.5 Models Comparison
- 5.6 Evaluation Summary
- Applications in X-ray Imaging
- 6.1 Motion Analysis
- 6.2 Temporal Changes Detection
- 6.3 Tracking
- Summary and Outlook
- 7.1 Conclusions
- 7.2 Further work
- List of Publications