Use this URL to cite or link to this record in EThOS:
Title: Image segmentation based on water flow analogy
Author: Liu, Xin
ISNI:       0000 0004 2668 3460
Awarding Body: University of Southampton
Current Institution: University of Southampton
Date of Award: 2009
Availability of Full Text:
Access from EThOS:
Access from Institution:
Segmenting objects with complex shapes, like segmenting vessels in iridology and detecting roads in remote sensing, is of practical significance in image analysis. Region growing and snakes are the main methods used in the field, but the former cannot yield an exact result whilst the latter has difficulties with topological changes. This thesis presents a new method, based on the paradigm of water flow. The mechanism embodies the fluidity of water and thus can deal with complicated objects. A snake-like force functional is utilized, including edge-based and region-based forces, guiding extraction to select the target feature. Water characteristics such as surface tension and adhesion are also implemented so that the smoothness of extracted contour and ability of flow to narrow branches are obtained. Furthermore, force field theories are incorporated for an alternative definition of the water flow forces so that the water flow framework becomes more generalised and flexible and the underlying theoretical basis is more consistent. Because of the 3-D nature of the physical water flow process, the extension of the model to higher dimensions is straightforward and has been implemented in this research. The higher dimensionality, however, increases the computational cost. To improve the efficiency, a water cooling system is proposed. This system identifies water elements that are inactive in previous flow iterations and removes them from the subsequent computations to reduce the total computational burden. The water cooling system provides an adjustable efficiency-controlling-facility and is also useful in 2-D applications. The new technique has been assessed on both synthetic and natural images. The topological adaptability and the geometrical flexibility of the model are assessed. In addition, the approach is shown to be able to segment shapes with weak contrast to their background. The good ability to handle noise, consistent with properties included in its formulation, has been justified both qualitatively and quantitatively. The effectiveness of the water cooling system is also proved graphically and quantitatively. Further, the method is applied to medical imaging problems such as MRI image segmentation, vessel detection and medical volume segmentation. The results justify the great potential of the new technique in the real-world applications.
Supervisor: Nixon, Mark Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: QA75 Electronic computers. Computer science