Welcome to my PhD thesis homepage. My name is Carlos Leung and I am currently completing my PhD research in the field of Computer Vision and Image Analysis at the University of Queensland, under the supervision of Prof. Brian C. Lovell from the University of Queensland and Dr. Changming Sun from CSIRO Mathematical and Information Sciences. The purpose of this website is to provide supplementary images and videos for my PhD thesis. The codes for all of the algorithms described in the thesis are also intended to be open-sourced.
Carlos Leung, "Efficient Methods for 3D Reconstruction from Multiple Images", Ph.D. dissertation, University of Queensland, 2005. [pdf]
Abstract:
This thesis explores the problem of reconstructing a three-dimensional (3D)
scene given a set of images or image sequences of the scene. It describes
efficient methods for the 3D reconstruction of static and dynamic scenes from
stereo images, stereo image sequences, and images captured from multiple
viewpoints. Novel methods for image-based and volumetric modelling approaches to
3D reconstruction are presented, with an emphasis on the development of
efficient algorithms which produce high quality and accurate reconstructions.
For image-based 3D reconstruction a novel energy minimisation scheme, Iterated
Dynamic Programming, is presented for the efficient computation of strong local
minima of discontinuity-preserving energy functions. Coupled with a novel
morphological decomposition method and subregioning schemes for the efficient
computation of a narrowband matching cost volume, the minimisation framework is
applied to solve problems in stereo matching, stereo-temporal reconstruction,
motion estimation, 2D image registration and 3D image registration. This thesis
establishes Iterated Dynamic Programming as an efficient and effective energy
minimisation scheme suitable for computer vision problems which involve finding
correspondences across images.
For 3D reconstruction from multiple view images with arbitrary camera placement,
a novel volumetric modelling technique, Embedded Voxel Colouring, is presented
that efficiently embeds all reconstructions of a 3D scene into a single output
in a single scan of the volumetric space under exact visibility. An adaptive
thresholding framework is also introduced for the computation of the optimal set
of thresholds to obtain high quality 3D reconstructions. This thesis establishes
the Embedded Voxel Colouring framework as a fast, efficient and effective method
for 3D reconstruction from multiple view images.
Organisation of the thesis and links to
supplementary images and videos:
Chapter 2: This chapter presents the preliminary background theory of
this thesis. A brief introduction to multiple view geometry will be presented,
describing in particular camera modelling, epipolar and projective geometry.
Useful algorithms for efficient computations such as the box filtering
technique, grey-scale morphology and the dynamic programming algorithm is also
presented.
Chapter 3: This chapter investigates the efficient computation of
matching costs. In problems which involve the search for matching
correspondences such as stereo reconstruction, it is necessary to compute the
matching likelihood for all candidate matches. This chapter presents two
algorithms for the efficient computation of matching costs within a narrowband.
Chapter 4:
This chapter explores efficient algorithms for energy minimisation. It presents
the computation of the stereo reconstruction and correspondence matching problem
by casting it into an energy minimisation framework. A novel energy minimisation
algorithm is described and results are presented to demonstrate its efficiency
and high quality solutions obtained. A derivation of the energy function to be
minimised is also presented, along with a detailed analysis of the research
literature in this field. This chapter also presents the efficient computation
of dynamic programming using grey-scale morphology.
Video simulation of the IDP optimisation process: IDP_minimisation_process.avi (400kB)
Chapter 5: This chapter presents the application of the novel energy minimisation scheme to efficiently solve a variety of computer vision and imaging problems. The novel energy minimisation framework is applicable to problems which involve finding correspondences, and can therefore be applied to solve problems such as stereo reconstruction, stereo-temporal reconstruction, motion estimation, 2D and 3D image registration problem. A review of literature for each of these application areas will be presented, and the application of the minimisation framework to solve these problems will be described.

Stereo-temporal reconstruction example 1: Stereo_Temporal.avi (200kB)
Stereo-temporal reconstruction example 2: Stereo_Temporal_Compare.avi (1MB)
2D breast registration results: left_view right_view realigned_right_view
3D brain registration results: left_slice_view right_slice_view realigned_right_view
left_3D_view right_3D_view realigned_3D_view
3D brain input volume: brain_input_slices.mov (27MB) brain_3D_rotation.mov (4MB)
Chapter 6: This chapter explores efficient methods for 3D reconstruction using a volumetric modelling approach. Volumetric modelling methods are particularly useful for multiple views reconstruction and occlusion modelling. Its disadvantages however are the expensive computational time required and the lack of spatial regularisation. This chapter presents a novel volumetric reconstruction algorithm. Three key properties necessary for volumetric reconstruction are identified and a literature review of volumetric modelling methods will be presented. An adaptive thresholding method will also be described to select the optimal set of threshold values.

Volumetric 3D reconstruction via EVC and globally minimal surfaces:
EVC_3D_Reconstruction.avi (1MB)
The
embedded volume computed encapsulates the reconstructions for all thresholds:
EVC_embedded_volume.avi (600kB)
Chapter 7: This chapter summarises the thesis and review the contributions. Directions for future research will also be explored.
List of Related Publications:
Carlos Leung, "3D
Dynamic Scene Reconstruction from Multi-View Image Sequences", Ph.D.
Confirmation Report, University of Queensland, 2003.
Carlos Leung and Brain C. Lovell, "3D
Reconstruction through Segmentation of Multi-View Image Sequences", APRS
Workshop on Digital Image Computing, pp. 87-92, Brisbane, Australia, 2003.
Carlos Leung, Ben Appleton and Changming Sun, "Embedded
Voxel Colouring", Digital Image Computing - Techniques and Applications,
C. Sun, H. Talbot, S. Ourselin and T. Adriaansen, Eds., vol. 2, pp. 97-106,
Sydney, Australia, December 2003.
Carlos Leung, Ben Appleton, Brian C. Lovell and Changming Sun, "An Energy
Minimisation Approach to Stereo-Temporal Dense Reconstruction",
International Conference on Pattern Recognition, vol. 4, pp. 72-75,
Cambridge, United Kingdom, 23-26 August, 2004.
Carlos Leung, Ben Appleton and Changming Sun, "Fast Stereo Matching by Iterated
Dynamic Programming and Quadtree Subregioning", British Machine Vision
Conference, A. Hoppe, S. Barman and T. Ellis, Eds., vol. 1, pp. 97-106,
Kingston, United Kingdom, 7-9 September, 2004..
This page was last updated on 06/10/2005.
There have been
visitors since 06/10/2005