No description available.

No description available.

The research discussed in this thesis is concerned with surface understanding from the viewpoint of recognition-oriented, scale-related processing based on surface curvatures and diffusion smoothing. Four problems below high level visual processing are investigated: 1) 3-dimensional data smoothing using a diffusion process; 2) Behaviour of shape features across multiple scales, 3) Surface segmentation over multiple scales; and 4) Symbolic description of surface features at multiple scales. In this thesis, the noisy data smoothing problem is treated mathematically as a boundary value problem of the diffusion equation instead of the well-known Gaussian convolution, In such a way, it provides a theoretical basis to uniformly interpret the interrelationships amongst diffusion smoothing, Gaussian smoothing, repeated averaging and spline smoothing. It also leads to solving the problem with a numerical scheme of unconditional stability, which efficiently reduces the computational complexity and preserves the signs of curvatures along the surface boundaries. Surface shapes are classified into eight types using the combinations of the signs of the Gaussian curvature K and mean curvature H, both of which change at different scale levels. Behaviour of surface shape features over multiple scale levels is discussed in terms of the stability of large shape features, the creation, remaining and fading of small shape features, the interaction between large and small features and the structure of behaviour of the nested shape features in the KH sign image. It provides a guidance for tracking the movement of shape features from fine to large scales and for setting up a surface shape description accordingly. A smoothed surface is partitioned into a set of regions based on curvature sign homogeneity. Surface segmentation is posed as a problem of approximating a surface up to the degree of Gaussian and mean curvature signs using the depth data alone How to obtain feasible solutions of this under-determined problem is discussed, which includes the surface curvature sign preservation, the reason that a sculptured surface can be segmented with the KH sign image alone and the selection of basis functions of surface fitting for obtaining the KH sign image or for region growing. A symbolic description of the segmented surface is set up at each scale level. It is composed of a dual graph and a geometrical property list for the segmented surface. The graph describes the adjacency and connectivity among different patches as the topological-invariant properties that allow some object's flexibility, whilst the geometrical property list is added to the graph as constraints that reduce uncertainty. With this organisation, a tower-like surface representation is obtained by tracking the movement of significant features of the segmented surface through different scale levels, from which a stable description can be extracted for inexact matching during object recognition.

by Or Siu Hang. / Thesis (M.Phil.)--Chinese University of Hong Kong, 1991. / Bibliography: leaves [102]-[105]. / Acknowledgements --- p.V / Chapter Chapter 1 --- Introduction / Chapter 1.1 --- The problem --- p.1 / Chapter 1.1.1 --- The correspondence problem --- p.5 / Chapter 1.1.2 --- The problem of surface reconstruction --- p.6 / Chapter 1.2 --- Our goal --- p.8 / Chapter 1.3 --- Previous works --- p.8 / Chapter 1.3.1 --- Constraints on matching --- p.10 / Chapter 1.3.2 --- Interpolation of disparity surfaces --- p.12 / Chapter Chapter 2 --- Preprocessing of images / Chapter 2.1 --- Which operator to use --- p.14 / Chapter 2.2 --- Directional zero-crossing --- p.14 / Chapter 2.3 --- Laplacian of Gaussian --- p.16 / Chapter 2.3.1 --- Theoretical background of the Laplacian of Gaussian --- p.18 / Chapter 2.3.2 --- Implementation of the operator --- p.21 / Chapter Chapter 3 --- Disparity Layers Generation / Chapter 3.1 --- Geometrical constraint --- p.23 / Chapter 3.2 --- Basic idea of disparity layer --- p.26 / Chapter 3.3 --- Consideration in matching --- p.28 / Chapter 3.4 --- effect of vertical misalignment of sensor --- p.37 / Chapter 3.5 --- Final approach --- p.39 / Chapter Chapter 4 --- Disparity combination / Chapter 4.1 --- Ambiguous match from different layers --- p.52 / Chapter 4.2 --- Our approach --- p.54 / Chapter Chapter 5 --- Generation of dense disparity map / Chapter 5.1 --- Introduction --- p.58 / Chapter 5.2 --- Cooperative computation --- p.58 / Chapter 5.2.1 --- Formulation of oscillation algorithm --- p.59 / Chapter 5.3 --- Interpolation by Gradient descent method --- p.69 / Chapter 5.3.1 --- Formulation of constraints --- p.70 / Chapter 5.3.2 --- Gradient projection interpolation algorithm --- p.72 / Chapter 5.3.3 --- Implementation of the algorithm --- p.78 / Chapter Chapter 6 --- Conclusion --- p.89 / Reference / Appendix (Dynamical behavior of the cooperative algorithm)

The overall argument this thesis makes is that topological object structures captured within hierarchical image descriptions are invariant to depictive styles and offer a level of abstraction found in many modern abstract artworks. To show how object structures can be extracted from images, two hierarchical image descriptions are proposed. The first of these is inspired by perceptual organisation; whereas, the second is based on agglomerative clustering of image primitives. This thesis argues the benefits and drawbacks of each image description and empirically show why the second is more suitable in capturing object strucutures. The value of graph theory is demonstrated in extracting object structures, especially from the second type of image description. User interaction during the structure extraction process is also made possible via an image hierarchy editor. Two applications of object structures are studied in depth. On the computer vision side, the problem of object classification is investigated. In particular, this thesis shows that it is possible to classify objects regardless of their depictive styles. This classification problem is approached using a graph theoretic paradigm; by encoding object structures as feature vectors of fixed lengths, object classification can then be treated as a clustering problem in structural feature space and that actual clustering can be done using conventional machine learning techniques. The benefits of object structures in computer graphics are demonstrated from a Non-Photorealistic Rendering (NPR) point of view. In particular, it is shown that topological object structures deliver an appropriate degree of abstraction that often appears in well-known abstract artworks. Moreover, the value of shape simplification is demonstrated in the process of making abstract art. By integrating object structures and simple geometric shapes, it is shown that artworks produced in child-like paintings and from artists such as Wassily Kandinsky, Joan Miro and Henri Matisse can be synthesised and by doing so, the current gamut of NPR styles is extended. The whole process of making abstract art is built into a single piece of software with intuitive GUI.

Driving Assistance Systems such as lane departure and front collision warning has caught great attention for its promising usage on road driving. This, this research focus on implementing lane departure and front collision warning at same time. In order to make the system really useful for real situation, it is critical that the whole process could be near real-time. Thus we chose Hough Transform as the main algorithm for detecting lane on the road. Hough Transform is used for that it is a very fast and robust algorithm, which makes it possible to execute as many frames as possible per frames. Hough Transform is used to get boundary information, so that we could decide if the car is doing lane departure based on the car's position in lane. Later, we move on to use front car's symmetry character to do front car detection, and combine it with Camshift tracking algorithm to fill the gap for failure of detection. Later we introduce camera calibration, stereo calibration, and how to calculate real distance from depth map.

Visual object recognition is one of the key human capabilities that we would like machines to have. The problem is the following: given an image of an object (e.g. someone's face), predict its label (e.g. that person's name) from a set of possible object labels. The predominant approach to solving the recognition problem has been to learn a discriminative model, i.e. a model of the conditional probability $P(l|v)$ over possible object labels $l$ given an image $v$. Here we consider an alternative class of models, broadly referred to as \emph{generative models}, that learns the latent structure of the image so as to explain how it was generated. This is in contrast to discriminative models, which dedicate their parameters exclusively to representing the conditional distribution $P(l|v)$. Making finer distinctions among generative models, we consider a supervised generative model of the joint distribution $P(v,l)$ over image-label pairs, an unsupervised generative model of the distribution $P(v)$ over images alone, and an unsupervised \emph{reconstructive} model, which includes models such as autoencoders that can reconstruct a given image, but do not define a proper distribution over images. The goal of this thesis is to empirically demonstrate various ways of using these models for object recognition. Its main conclusion is that such models are not only useful for recognition, but can even outperform purely discriminative models on difficult recognition tasks. We explore four types of applications of generative/reconstructive models for recognition: 1) incorporating complex domain knowledge into the learning by inverting a synthesis model, 2) using the latent image representations of generative/reconstructive models for recognition, 3) optimizing a hybrid generative-discriminative loss function, and 4) creating additional synthetic data for training more accurate discriminative models. Taken together, the results for these applications support the idea that generative/reconstructive models and unsupervised learning have a key role to play in building object recognition systems.

Visual object recognition is one of the key human capabilities that we would like machines to have. The problem is the following: given an image of an object (e.g. someone's face), predict its label (e.g. that person's name) from a set of possible object labels. The predominant approach to solving the recognition problem has been to learn a discriminative model, i.e. a model of the conditional probability $P(l|v)$ over possible object labels $l$ given an image $v$. Here we consider an alternative class of models, broadly referred to as \emph{generative models}, that learns the latent structure of the image so as to explain how it was generated. This is in contrast to discriminative models, which dedicate their parameters exclusively to representing the conditional distribution $P(l|v)$. Making finer distinctions among generative models, we consider a supervised generative model of the joint distribution $P(v,l)$ over image-label pairs, an unsupervised generative model of the distribution $P(v)$ over images alone, and an unsupervised \emph{reconstructive} model, which includes models such as autoencoders that can reconstruct a given image, but do not define a proper distribution over images. The goal of this thesis is to empirically demonstrate various ways of using these models for object recognition. Its main conclusion is that such models are not only useful for recognition, but can even outperform purely discriminative models on difficult recognition tasks. We explore four types of applications of generative/reconstructive models for recognition: 1) incorporating complex domain knowledge into the learning by inverting a synthesis model, 2) using the latent image representations of generative/reconstructive models for recognition, 3) optimizing a hybrid generative-discriminative loss function, and 4) creating additional synthetic data for training more accurate discriminative models. Taken together, the results for these applications support the idea that generative/reconstructive models and unsupervised learning have a key role to play in building object recognition systems.

Thesis (M.Sc.)--York University, 2007. Graduate Programme in Computer Science. / Typescript. Includes bibliographical references (leaves118-124). Also available on the Internet. MODE OF ACCESS via web browser by entering the following URL: http://gateway.proquest.com/openurl?url_ver=Z39.88-2004&res_dat=xri:pqdiss&rft_val_fmt=info:ofi/fmt:kev:mtx:dissertation&rft_dat=xri:pqdiss:MR38774

Thesis (Ph. D.)--University of Hong Kong, 2010. / Includes bibliographical references (p. 139-144). Also available in print.