2003 | 2002 | 2001 | 2000 | 1999...
2003
Name: Macrini, Diego A.
Supervisor(s): Dickinson, Sven
Degree: M.Sc.
Date: July, 2003
Title: Indexing and Matching for View-Based 3-D Object Recognition Using Shock Graphs
Abstract: A shock graph is a shape abstraction that decomposes a shape into hierarchically organized primitive parts. In this thesis, we propose a novel shock graph computation algorithm that yields stable graphs under noise and shape deformation due to viewpoint changes. Next, we extend the indexing and matching framework for hierarchical structures introduced by Shokoufandeh et al., and apply it to the problem of matching shock graphs representing views of 3-D objects. The indexing performance is improved by a vote accumulation algorithm that efficiently solves a multiple one-to-one assignment of votes. In turn, the matching algorithm is extended by ensuring the satisfaction of all the hierarchical constraints encoded in the graphs. We show that the proposed integrated framework is effective in recognizing object shapes and in estimating the pose of their corresponding 3-D object models. We demonstrate the performance of the framework with a database of 2688 object views.
Name: El-Maraghi, Thomas F.
Supervisor(s): Jepson, Allan
Degree: Ph.D.
Date: February, 2003
Title: Robust on-line appearance models for visual tracking
A framework for learning robust, adaptive appearance models to be used for motion-based tracking of natural objects is proposed. The approach involves a mixture of stable image structure, learned over long time courses, along with 2-frame transient motion information and an outlier process. This class of appearance models is shown to have low storage requirements and that the model parameters can be learned efficiently with an on-line variant of the expectation-maximization (EM) algorithm. Two implementations of appearance models based on the framework presented here are developed. The first is based on the filter responses from a steerable pyramid. This model is used in a motion-based tracking algorithm to provide robustness in the face of image outliers, such as those caused by occlusions. It also provides the ability to adapt to natural changes in appearance, such as those due to facial expressions or variations in 3D pose. The second implementation is based on a robust representation of the color of the target object. Experimental results on a variety of natural image sequences of people moving within cluttered environments are shown, both for the wavelet- and color-based appearance models individually and in combination with one another. A quantitative analysis of the performance of the tracker is provided for one of the test sequences.
2002
Name: Hasinoff, Sam
Supervisor(s): Kutulakos, Kyros
Degree: M.Sc.
Date: October, 2002
Title: Three-Dimensional Reconstruction of Fire from Images
Abstract: Interesting visual simulations have been proposed for physical phenomena such as fire and smoke, but scant attention has been paid to reconstructing models of these phenomena from multiple images of real scenes. Here we study the problem of reconstructing three-dimensional models of fire from synchronized images or video sequences. Using a physically-based imaging model, we reduce the reconstruction of fire to the little-studied problem of computerized tomography with an extremely limited number of views (as few as two). We analyse the space of photo-consistent reconstructions, propose a novel reconstruction method for fire, and develop an image-based method for rendering the reconstructions. The reconstruction method involves parameterizing the fire with a linear superposition of Gaussians, estimating the density according to the input images, and applying a stochastic resampling scheme to refine the reconstruction. Experimental results are presented for a variety of synthetic and real fire datasets.
2001
Name: Midgley, John
Supervisor(s): Jepson, Allan
Degree: M.Sc.
Date: August, 2001
Title: Probabilistic Eigenspace Object Recognition in the Presence of Occlusion
Abstract: Eigenspace approaches to object recognition have achieved impressive results in constrained environments. In general, however, these techniques tend to be brittle, as they are not naturally robust to changes in object scale, affine object transformations in an image, object occlusion, or background clutter. In this thesis, we present a novel approach to handling occlusions and background clutter when using an eigenspace approach to object recognition. The method presented involves an intuitive probabilistic formulation that exploits local information intrinsic to the eigenspace model. In addition, we explore the usefullness of enforcing local image contraints during fitting. A general, efficient, intrinsically parallel, search procedure is proposed that uses the framework. Experimental results indicate that the technique can reliably handle much higher degrees of occlusion, in comparison to previous techniques.
Name: Estrada, Francisco
Supervisor(s): Jepson, Allan
Degree: M.Sc.
Date: March 2001
Title: Estimation of Surface Orientation from a Single Image
Abstract: It is known that perspective projection plays an important role in the human visual system. It enables an observer to obtain relevant information about the 3D structure of the world even when stereographic and focusing information are unavailable or of little use given the scale of the structures in the image. This thesis explores the determination of surface orientation from a single view using the perspective information provided by the orientation of line segments present in the image. The issues addressed are the extraction of relevant features, the extraction of the perspective information that encodes the orientation of planar surfaces, and the grouping of features into areas which have a consistent orientation. Experimental results in real world images are presented and possible extensions to this work are discussed.
2000
Name: Listgarten, Jennifer
Supervisor(s): Jepson, Allan
Degree: M.Sc.
Date: October 2000
Title: Exploring Qualitative Probabilities for Image Understanding
Abstract: In this thesis we explore and work with a particular probabilistic framework for image interpretation called Qualitative Probabilities. Introduced by Jepson and Mann, Qualitative Probabilities formalize the notion of non-accidentalness, a cornerstone of object recognition.
First we examine the search space associated with Qualitative Probabilities. We also experimentally verify one of the underlying principles of the theory, the asymptotic rate of 'accidents'. Then we incorporate Qualitative Probabilities into a relatively simple search which we find to be efficient and effective. Comparing search for interpretations using Qualitative Probabilities to search using a more standard 'cover' measure, we find that the former is far superior both in terms of efficiency and quality of block models found. Lastly, we design and test a new search algorithm, called Cascade search, that uses Qualitative Probabilities.
Name: Qureshi, Faisal
Supervisor(s): Terzopoulos, Demetri
Degree: M.Sc.
Date: January 2000
Title: Constructing Anatomically Accurate Face Models using Computed Tomography and Cyberware data
Abstract: Facial animation and cranio-facial surgery simulation both stand to benefit from the development of anatomically accurate computer models of the human face. State-of-the-art biomechanical models of the face have shown promise in animation, but they are inadequate for the purposes of cranio-facial surgery simulation. The goal of this thesis is to develop an improved facial model, using Cyberware data which captures the external structure and appearance of the face and head, as well as computed tomography (CT) data which captures the internal structure of facial soft and hard tissues. To this end, we develop algorithms to (1) register the CT and Cyberware datasets, (2) extract from the CT data a skull subsurface which serves as a foundation of the soft-tissue model, and (3) compute thickness of facial soft-tissues over the face.
