Communication and Information Technology Research - Technical Reports (1997-2007)
Permanent URI for this collectionhttps://hdl.handle.net/2292/2623
Browse
Recent Submissions
Item Finite Difference Methods for Linear Shape from Shading(CITR, The University of Auckland, New Zealand, 2001) Wei, Tiangong; Klette, ReinhardThis paper presents and analyzes four finite difference methods for linear shape from shading problem. Comparisons of accuracy, solvability, stability and convergence of these methods indicate that the weighted semi-implicit method and the box method are better than the other ones because they are easily calculated, more accurate, faster in convergence and unconditionally stable.Item Height from Gradient with Surface Curvature and Area Constraints(CITR, The University of Auckland, New Zealand, 2001) Wei, Tiangong; Klette, ReinhardThis paper presents a new algorithm for height from gradients and analyzes its performance. To derive this algorithm, we combine the integrability constraint and the surface curvature and area constraints into a single functional, which is then minimized. Therefore, the changes of height maps will be more regular. The Frankot-Chellappa-algorithm is a special case of our algorithm in the sense that it uses a subset of constraints only.Item Navigation Using Optical Flow Fields: An Application of Dominant Plane Detection(CITR, The University of Auckland, New Zealand, 2002) Kawamoto, Kazuhiko; Yamada, Daisuke; Klette, ReinhardVideo sequences capturing real scenes may be interpreted with respect to a dominant plane which is a planar surface covering 'a majority' of pixels in an image of a video sequence, i.e. that planar surface which is represented in the image by a maximum number of pixels. In this paper, we propose an algorithm for the detection of dominant planes from optical flow fields caused by camera motion in a static scene. We, in particular, intend to adopt this algorithm as a module for obstacle detection in vision-based navigation of autonomous robots.Item Skeletons in Digital Image Processing(CITR, The University of Auckland, New Zealand, 2002) Klette, GiselaSkeletonization is a transformation of a component of a digital image into a subset of the original component. There are different categories of skeletonization methods: one category is based on distance transforms, and a specified subset of the transformed image is a distance skeleton. The original component can be reconstructed from the distance skeleton. Another category is defined by thinning approaches; and the result of skeletonization using thinning algorithms should be a connected set of digital curves or arcs. Motivations for interest in skeletonization algorithms are the need to compute a reduced amount of data or to simplify the shape of an object in order to find features for recognition algorithms and classifications. Additionally the transformation of a component into an image showing essential characteristics can eliminate local noise at the frontier. Thinning algorithms are a very active area of research, with a main focus on connectivity preserving methods allowing parallel implementation. There are hundreds of publications on different aspects of these transformations. This report reviews contributions in this area with respect to properties of algorithms and characterizations of simple points, and informs about a few new results.Item Characterizations of Simple Pixels in Binary Images(CITR, The University of Auckland, New Zealand, 2002) Klette, GiselaThere are hundreds of publications on different aspects of image deformations that preserve topological properties of the original binary image. The notion of a "simple pixel" is of fundamental importance for these transformations. Simple pixels in 2D binary images have been characterized in a number of ways. This paper reviews some of these characterizations and it points out that some of them are actually equivalent.Item Evaluation of an Adaptive Composite Gaussian Model in Video Surveillance(CITR, The University of Auckland, New Zealand, 2002) Zang, Qi; Klette, ReinhardVideo surveillance systems seek to automatically identify events of interest in a variety of situations. Extracting a moving object from background is the most important step of the whole system. There are many approaches to track moving objects in a video surveillance system. These can be classified into three main groups: feature-based tracking, background subtraction, and optical flow techniques. Background subtraction is a region-based approach where the objective is to identify parts of the image plane that are significantly different to the background. In order to avoid the most common problems introduced by gradual illumination changes, waving trees, shadows, etc., the background scene requires a composite model. A mixture of Gaussian distributions is most popular. In this paper, we classify and discuss several recently proposed composite models. We have chosen one of these for implementation and evaluate its performance. We also analyzed its benefits and drawbacks, and designed an improved version of this model based on our experimental evaluation. One stationary camera has been used.Item Cylindrical Panoramic Cameras - From Basic Design to Applications(CITR, The University of Auckland, New Zealand, 2002) Huang, Fay; Kang, Shou; Gimel'farb, Georgy; Reulke, Ralf; Scheele, Martin; Scheibe, KarstenThis paper reviews major steps in designing, producing and applying cylindrical panoramic cameras. Basically such a camera is characterized by rotating linear sensors capturing one image column at a time. Extremely high numbers of pixels per line allow capturing of super-high resolution panoramic images. These can be used for stereo visualisation and stereo reconstruction, as will be discussed in this paper.Item Regularization Method for Depth from Noisy Gradient Vector Fields(CITR, The University of Auckland, New Zealand, 2002) Wei, Tiangong; Klette, ReinhardThis paper presents a regularization method for surface reconstruction from noisy gradient vector fields. The algorithm takes as its input a discrete gradient vector field, obtained by applying a Shape from Shading or Photometric Stereo method. To derive this algorithm, we combine the integrability constraint and the surface curvature and area constraints into a single functional, which is then minimized. Therefore, value changes in the height or depth map will be more regular. To solve the minimization problem, we employ the Fourier transform theory rather than the Variational Principle. The Fourier transform of the (unknown) surface is expressed as a function of the (given) gradient's Fourier transforms. The relative depth values can be obtained by an inverse Fourier Transform and by choosing associated weighting parameters. The method is evaluated on gradient data delivered by a Photometric Stereo algorithm.Item External versus Internal Parameterizations for Lengths of Curves with Nonuniform Samplings(CITR, The University of Auckland, New Zealand, 2002) Kozera, Ryszard; Noakes, LyleThis paper studies differences in estimating length (and also trajectory) of an unknown parametric curve from an ordered collection of data points qi = GAMA (ti), with either the ti's known or unknown. For the ti's uniform (known or unknown) piecewise Lagrange interpolation provides efficient length estimates, but in other cases it may fail. In this paper, we apply this classical algorithm when the ti's are sampled according to first ALPHA-order and then when sampling is EPSILON-uniform. The latter was introduced in [20] for the case where ti's are unknown. In the present paper we establish new results for the case when the ti's are known for both types of samplings. For curves sampled EPSILON-uniformly comparison is also made between the cases, where the tabular parameters ti's are known and unknown. Numerical experiments are carried out to investigate sharpness of our theoretical results. The work may be of interest in computer vision and graphics, approximation and complexity theory, digital and computational geometry, and digital image analysis.Item Recovery of Coloured Surface Reflectances Using the Photometric Stereo Method(CITR, The University of Auckland, New Zealand, 2002) Chen, Chia-Yen; Klette, Reinhard; Chen, Chi-FaWe discuss a fast and efficient method to estimate the surface reflectance values with respect to different wavelength of light. Qualitative and quantitative evaluations show that surfaces rendered using surface reflectances obtained by the proposed method are more realistic when compared with conventional rendering methods.Item Digital flatness and related combinatorial problems(CITR, The University of Auckland, New Zealand, 2002) Brimkov, ValentinIn the present notes we define and study the notion of digital flatness. We extend to two dimensions various definitions and results about digital straightness, in particular, we resolve a conjecture of M. Nivat for the case of digital planes and define and characterize 2D Sturmian rays.Item Combinatorics on Incidence Pseudographs(CITR, The University of Auckland, New Zealand, 2003) Klette, ReinhardIncidence pseudographs (or the dual model of complexes based on a bounded-by relation) are popular models for pixels or voxels and relations between these. They allow a definition of a topological space, and combinatorial formulas characterize open and closed sets in this topology. K. Voss characterized in 1993 open regions in incidence pseudographs. This article reviews these results and provides also combinatorial formulas for closed regions.Item A Review on Research and Applications of Cylindrical Panoramas(CITR, The University of Auckland, New Zealand, 2003) Klette, Reinhard; Gimel'farb, Georgy; Huang, Fay; Scheibe, Karsten; Scheele, Martin; Börner, AnkoThis paper reviews research related to the design, production and application of cylindrical panoramic cameras. Such a camera is characterized by rotating linear sensors capturing one image column at a time. This allows for accurate mappings onto a cylindrical image surface and very high image resolutions paid by motion distortions in dynamic scenes. These panoramic images can be used, for example, for stereo visualization and stereo reconstruction in applications where extremely high image resolution is of benefit (for static scenes). The paper deals especially with aspects of stereo visualization and reconstruction.Item A Comparative Discussion of Distance Transforms and Simple Deformations in Image Processing(CITR, The University of Auckland, New Zealand, 2003) Klette, GiselaAlgorithms for transformations of digital images into reconstructible subsets of the original image, and algorithms for deformations of digital images into topologically equivalent images are subjects of hundreds of publications. Two images are topologically equivalent if their adjacency trees are isomorphic. Skeletonization is a transformation of components of a digital image into a subset of the original component. There are different categories of skeletonization methods: one category is based on distance transforms, and a specified subset of the transformed image is called distance skeleton. The original component can be reconstructed from the distance skeleton. But the result is not a topologically equivalent image. A different category is defined by thinning approaches, and the result of skeletonization using thinning algorithms should be a topologically equivalent image. Thinning algorithms are one-way simple deformations, which mean object points are transferred into background points without destroying the topology of the image. Two-way simple deformations transfer object points into background points and vice versa without destroying the topology of the image. This report reviews contributions in this area with respect to properties of algorithms and characterizations of simple points, and informs about a few new results.Item Connectivity of discrete planes(CITR, The University of Auckland, New Zealand, 2003) Brimkov, Valentin; Barneva, RenetaStudying connectivity of discrete objects is a major issue in discrete geometry and topology. In the present work we deal with connectivity of discrete planes in the framework of Reveill es analytical definition [11]. Accordingly, a discrete plane is a set P (a, b, c, ,,,,, ) of integer points (x, y, z) satisfying the Diophantine inequalities 0 ax + by + cz + . The parameter Z estimates the plane intercept while N is the plane thickness. Given three integers (plane coefficients) a, b, and c with 0 a b c, one can seek the maximal for which the discrete plane P (a, b, c, ,,,,, ) is disconnected. We call this remarkable topological invariable the connectivity number of P (a, b, c, ,,,,, ) and denote it (a, b, c). Despite several attempts over the last ten years to determine the connectivity number, this is still an open question. In the present paper we propose a solution to the problem. For this, we first investigate some combinatorial properties of discrete planes. These structural results facilitate the deeper understanding of the discrete plane structure. On this basis, we obtain a series of results, in particular, we provide an explicit solution to the problem under certain conditions. We also obtain exact upper and lower bounds on (a, b, c) and design an O(a log b) algorithm for its computation.Item Accurately Measuring the Size of the Pupil of the Eye(CITR, The University of Auckland, New Zealand, 2003) Lin, Xiang; Craig, Jennifer; Klette, Gisela; Klette, ReinhardA new method is developed to accurately analyze the elliptical character of the contour of human s pupil in images captured by a special infrared CCD video camera. The method is an effective combination of edge detection algorithms, labelling method and least square fitting technique, which makes it robust to the eyes of different human races and easy to trace the random movements of the pupil. It even solved the problem of Purkinje phenomenal in images, which was previously only avoided by adjusting the light direction [1]. The parameters of a fitted ellipse, including the length of both axes, the orientation and the center position, are output as result for further analysis.Item Multigrid Analysis of Curvature Estimators(CITR, The University of Auckland, New Zealand, 2003) Hermann, Simon; Klette, ReinhardThis report explains a new method for the estimation of curvature of plane curves and compares it with a method which has been presented in [2]. Both methods are based on global approximations of tangents by digital straight line segments. Experimental studies show that a replacement of global by local approximation results in errors which, in contrast to the global approximation, converge to constants > 0. We also apply the new global method for curvature estimation of curves to surface curvature estimation, and discuss a method for estimating mean curvature of surfaces which is based on Meusnier's theorem.Item Object Classification and Tracking in Video Surveillance(CITR, The University of Auckland, New Zealand, 2003) Zang, Qi; Klette, ReinhardThe design of a video surveillance system is directed on automatic identification of events of interest, especially on tracking and classification of moving vehicles or pedestrians. In case of any abnormal activities, an alert should be issued. Normally a video surveillance system combines three phases of data processing: moving object extraction, moving object recognition and tracking, and decisions about actions. The extraction of moving objects, followed by object tracking and recognition, can often be defined in very general terms. The final component is largely depended upon the application context, such as pedestrian counting or traffic monitoring. In this paper, we review previous research on moving object tracking techniques, analyze some experimental results, and finally provide our conclusions for improved performances of traffic surveillance systems. One stationary camera has been used.Item 3D Reconstruction Using Shape from Photometric Stereo and Contours(CITR, The University of Auckland, New Zealand, 2003) Chen, Chia-Yen; Klette, ReinhardIn this work, we further discuss an approach to 3D shape recovery by combining photometric stereo and shape from contours methods. Surfaces recovered by photometric stereo are aligned, adjusted and merged according to a preliminary 3D model obtained by shape from contours. Comparisons are conducted to evaluate the performances of different methods. It has been found that the proposed combination provides more accurate shape recovery than using either photometric stereo or shape from contours alone.Item Authoring and Visualizing Stereo Panoramic Images with Independent Objects(CITR, The University of Auckland, New Zealand, 2003) Chen, Johnson; Wei, Shou-KangObject embedded stereo panoramic images have received increasing attention recently due to their numerous applications in computer vision. This article explores an approach to embed object images, both dynamic and static, into background panoramas. The approach involves three steps; namely, image acquisition, post-acquisition, and image visualization. A GUI tool is developed for image registering, merging, and pose recovering. A panoramic player is also developed for visualizing the resultant anaglyphic panorama.