VVG05

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/1030

Browse

Now showing 1 - 20 of 33

Rigorous Computing in Computer Vision
(The Eurographics Association, 2005) Farenzena, Michela; Fusiello, Andrea; Mike Chantler
In this paper we discuss how Interval Analysis can be used to solve some problems in Computer Vision, namely autocalibration and triangulation. The crucial property of Interval Analysis is its ability to rigorously bound the range of a function over a given domain. This allows to propagate input errors with guaranteed results (used in multi-views triangulation) and to search for solution in non-linear minimisation problems with provably correct branch-and-bound algorithms (used in autocalibration). Experiments with real calibrated images illustrate the interval approach.
A Model-Based Approach to Image Relighting with a Potential for Real-Time Implementation
(The Eurographics Association, 2005) Madsen, Claus B.; Laursen, Rune; Mike Chantler
Image relighting is a very unique special visual effect which promises to have many important practical applications. Image relighting is essentially the process of, given one or more images of some scene, computing what that scene would look like under some other (arbitrary) lighting conditions, e.g., changing positions and colors of light sources. Image relighting can for example be used for interior light design. This paper describes an approach to image relighting which can be implemented to run in real-time by utilizing graphics hardware, as opposed to other state-of-the-art approaches which at best run at a few frames per second.
SVD-Matching using SIFT Features
(The Eurographics Association, 2005) Delponte, Elisabetta; Isgrò, Francesco; Odone, Francesca; Verri, Alessandro; Mike Chantler
The paper tackles the problem of feature points matching between pair of images of the same scene. This is a key problem in computer vision. Among the many possible applications of feature matching we are motivated for helping in the initialisation of a 3D registration procedure. The method we discuss here is a version of the SVD matching proposed by Pilu, modified in order to cope with large scale variations. We detail the algorithm and present experimental evidence of the improvement in performance. The main contribution of this work is in showing that this compact and easy algorithm can be used for large-baseline matching.
Visual Modeling: Unifying Graphics and Vision
(The Eurographics Association, 2005) Terzopoulos, Demetri; Mike Chantler
The computer vision and computer graphics fields have been developing largely independently since their genesis in the early 1960s. However, there is now a significant amount of exciting research at the intersection of graphics and vision, and it is bringing the two disciplines closer together. Since the mid 1980s, my visual modeling research has taken a unified approach to vision and graphics, treating them formally as mutually converse problems. In this talk, I will first review early work on deformable models for image synthesis and analysis and show how these physics-based models ushered in a new paradigm known as image-based modeling. I will also present our recent work on image-based rendering, which again spans vision and graphics. Even more provocative opportunities for unifying graphics and vision are motivated by sophisticated, biology-based models: Within an artificial life paradigm, we have developed comprehensive computational models of humans and lower animals that take into consideration the relevant anatomy, biomechanics, and cognitive science. Central to these models is their ethological constituent, which is driven by active vision within the dynamic virtual environment. In particular, our current work on virtual humans focuses on the lifelike animation of visually perceptive, autonomous pedestrians in urban environments through the integration of (reactive) behavioral and (deliberative) cognitive components. Our work also furthers the cause of exploiting visually and behaviorally realistic virtual worlds for the development and testing of machine vision systems. To this end, I will demonstrate a surveillance system in a virtual train station populated by our autonomous virtual pedestrians. The system features a sensor network of readily recon- figurable active virtual cameras that generate synthetic video feeds emulating those generated by real surveillance cameras monitoring public spaces. Such research would be more or less infeasible in the real world in view of the effort and cost of deploying, modifying, and experimenting with an appropriately extensive camera network in a public space the size of a train station.
Depth Enhanced Panoramas
(The Eurographics Association, 2005) Bahmutov, Gleb; Popescu, Voicu; Mudure, Mihai; Sacks, Elisha; Mike Chantler
Depth enhanced panoramas are a practical solution to the difficult problem of inside-looking-out modeling. Depth enhanced panoramas extend color panoramas to support viewpoint translation, while retaining their speed, convenience, and low cost. Depth enhanced panoramas are built incrementally from same-center-of-projection dense-color and sparse-depth frames that are acquired, registered, and merged at the rate of 5 frames per second. The evolving depth enhanced panorama is rendered continually to provide immediate operator feedback. The viewpoint translation range is increased by combining multiple depth enhanced panoramas in real time. Depth enhanced panoramas are combined using either a splat-based, disconnected representation, or, at pixel level, using a GPU fragment program. Depth enhanced panoramas are built in minutes with $3,000 hardware and support photorealistic rendering of complex room-sized environments.
Adaptive Grid Optical Tomography
(The Eurographics Association, 2005) Ihrke, Ivo; Magnor, Marcus; Mike Chantler
Image-based modeling of semi-transparent, dynamic phenomena is a challenging task. We present an optical tomography method that uses an adaptive grid for the reconstruction of a three-dimensional density function from its projections. The proposed method is applied to reconstruct thin smoke and flames volumetrically from synchronized multi-video recordings. Our adaptive reconstruction algorithm computes a time-varying volumetric model, that enables the photorealistical rendering of the recorded phenomena from arbitrary viewpoints. In contrast to previous approaches we sample the underlying unknown, three-dimensional density function adaptively which enables us to achieve a higher effective resolution of the reconstructed models.
Motion Analysis in Video: Dolls, Dynamic Cues and Modern Art
(The Eurographics Association, 2005) Collomosse, J. P.; Hall, P. M.; Mike Chantler
This paper addresses the problem of synthesising animations from video clips; in particular emphasising the motion of tracked objects. We introduce "dynamic cues" as a class of motion emphasis cue, encompassing traditional animation techniques such as anticipation and exaggeration. We present methods for automatically synthesising such cues within video premised upon the recovery of articulated figures, and the subsequent manipulation of the recovered pose trajectories. Additionally, we apply our motion emphasis framework to emulate artwork in the Futurist style, popularised by Duchamp.
Automatic Non-Photorealistic Rendering through Soft-Shading Removal: A Colour-Vision Approach
(The Eurographics Association, 2005) Olmos, A.; Kingdom, F. A. A.; Mike Chantler
This paper presents a non-photorealistic rendering algorithm that produces "stylised-style" images by removing the soft shading from the image and by giving objects extra definition through black outlines. The method of shading removal is based on a model of the architecture of the human colour vision system. Some image results are provided and the possible extension of the algorithm using a back-propagation neural network is discussed.
Sensitivity Analysis in Image Synthesis
(The Eurographics Association, 2005) Trujillo, M.; Izquierdo, E.; Mike Chantler
Intermediate virtual images are used in the evaluation of disparity estimations. The analysis is based on the effect of disparity inaccuracies in the relative quality of virtual images. The peak-signal-to-noise ratio and the percentage of visual errors are used to assess the quality of virtual images generated with distorted disparity fields. Computer simulation results show that the peak-signal-to-noise ratio is more affected by a small magnitude of perturbations two or three pixels than by a high frequency of perturbations 15% or 20%. However, the percentage of visual errors is more affected by the frequency of perturbations. By using distorted disparity fields, the deterioration in the quality of virtual images is imperceptible to the human eye.
Regularised Anisotropic Nonlinear Diffusion for Rendering Refraction in Volume Graphics
(The Eurographics Association, 2005) Rodgman, David; Chen, Min; Mike Chantler
Rendering refraction in volume graphics requires smoothly distributed normals to synthesise good quality visual representations. Such refractive visualisation is more susceptible to noise in the data than visualisations that do not involve refraction. In this paper, we addresses the need for improving the continuity of voxel gradients in discretely sampled volume datasets using nonlinear diffusion methods, which was originally developed for image denoising. We consider the necessity for minimising unnecessary geometrical distortion, detail the functional specification of a volumetric filter for regularised anisotropic nonlinear diffusion (R-ANLD), discuss the further improvements of the filter, and compare the efficacy of the filter with an anisotropic nonlinear diffusion (ANLD) filter as well as a Gaussian filter and a linear diffusion filter. Our results indicate that it is possible to make significant improvements in image quality in refractive rendering without excessive distortion.
Correspondenceless Stereo for 3-D Iris Location
(The Eurographics Association, 2005) Anderson, Tom; Trucco, Emanuele; Razeto, Marco; Mike Chantler
We present a correspondenceless stereo system locating the circular contour of an iris (limbus) in space, and therefore its 3-D plane. We avoid correspondence search completely by intersecting a bundle of conjugate epipolar lines with the elliptical images of the limbus in the stereo pair of images, which gives correspondences directly. The ellipses are located by active ellipse fitting. An efficient simulated annealing implementation achieves reliable iris location with uncontrolled illumination and eye or head movements. Tests with ground-truthed 3-D setups as well as real eye images indicate very good accuracy.
Use of Computer Vision/ Computer Graphics Collaboration for 3D Rotoscopy
(The Eurographics Association, 2005) Gagalowicz, André; Mike Chantler
This talk introduces 3D rotoscopy and its role in post-production applications, and the computer vision/computer graphics strategies used. The case of rigid objects is presented first, for which well-established approaches can be used. The case of articulated objects is considered next, with special attention to the case of a full human body tracking (ignoring significant shape deformations introduced by baggy clothes). Results related to the tracking of professional golfers and face tracking, a challenging case of deformable objects, will be presented. The talk will be concluded by a presentation of further, possible applications of research done at the MIRAGES laboratory at INRIA Rocquencourt.
Visual Recognition of Man-made Materials and Structures in an Office Environment
(The Eurographics Association, 2005) Song, Y. Z.; Town, C. P.; Mike Chantler
This paper demonstrates a new approach towards object recognition founded on the development of Neural Network classifiers and Bayesian Networks. The mapping from segmented image region descriptors to semantically meaningful class membership terms is achieved using Neural Networks. Bayesian Networks are then employed to probabilistically detect objects within an image by means of relating region class labels and their surrounding environments. Furthermore, it makes use of an intermediate level of image representation and demonstrates how object recognition can be achieved in this way.
Color Reduction by Using a new Self-Growing and Self-Organized Neural Network
(The Eurographics Association, 2005) Atsalakis, A.; Papamarkos, N.; Mike Chantler
A new method for the reduction of the number of colors in a digital image is proposed. The new method is based on the developed of a new neural network classifier that combines the advantages of the Growing Neural Gas (GNG) and the Kohonen Self-Organized Feature Map (SOFM) neural networks. We call the new neural network: Self-Growing and Self- Organized Neural Gas (SGONG). Its main advantage is that it defines the number of the created neurons and their topology in an automatic way. As a consecutive, isolated color classes, which may correspond to significant image details, can be obtained. The SGONG is fed by the color components and additional spatial features. To speed up the entire algorithm and to reduce memory requirements, a fractal scanning sub-sampling technique is used. The method is applicable to any type of color images and it can accommodate any type of color space.
A New Framework for Trademark Retrieval Based on Size Functions
(The Eurographics Association, 2005) Cerri, A.; Ferri, M.; Giorgi, D.; Mike Chantler
We propose a new, effective system for Content Based trademark retrieval, which involves Size Functions. Three different classes of shape descriptors are combined, for a total amount of 25 measuring functions. The evaluation has been performed on a database of 1182 trademark images, provided by the UK Patent Office.
Bitvectors for Robust Hierarchical Template Matching
(The Eurographics Association, 2005) Tweed, David; Mike Chantler
Many target detection problems involve objects where the primary variability in appearance is due to changes amongst characteristic configurations (as opposed to the systematic variability of object rotation or illumination changes). It is then important to utilise as much of the correlation between features as possible. Detecting pedestrians is such a problem and was tackled by Gavrila [Gav98] using a large set of exemplar templates combined with hierarchical matching via Distance Transforms. We describe a variant using a robust distance function and explicit allowance for occlusions. Our innovation is using bitwise logical operators to test against multiple exemplars in parallel.
Texture Mapping Volume Objects
(The Eurographics Association, 2005) Shen, P.; Willis, P.; Mike Chantler
We present a combination of image-based texture mapping and projective space (pseudo-solid) texture. This imagebased texture mapping is useful for objects defined from volume datasets. The paper makes three main contributions. First, it introduces the combination of the image-based two-part texture mapping and projective space texture mapping for volume objects. Second, it presents a multi-resolution technique to overcome problems with projecting at glancing angles and to eliminate artifacts due to the resolution limitations. Third, it presents the pixel-level data-dependent interpolation technique in projective image warping. The proposed approach leads to superior quality of texture and thus provides an optional solution for texturing volume objects.The results show the effectiveness and quality of rendered images.
Merging Graphics and Vision for 3D Face Recognition
(The Eurographics Association, 2005) Bai, Li; Song, Yi; Mike Chantler
This paper presents a new approach to automatic 3D face modelling from unstructured point cloud data. An efficient B-Spline surface-fitting algorithm is used to obtain an initial parametric surface for each face point cloud data set. Knot vectors for each individual face surface are then standardised to produce a set of uniform knot vectors so that all the surfaces can be seen as fitted with the same set of knot vectors. Mapping from object space to shape space can then be established so that each 3D face can be described by a small number of shape descriptors. The use of shape descriptors allows automatic registration between face models. More importantly, it allows dynamic facial variation to be modelled and analysed via 3D warping, resulting in a powerful approach to quantifying the differences among individuals required for face recognition. 3D warping is often used in simulations in computer graphics. This paper explains, for the first time, how 3D warping can be exploited for face recognition based on multi-resolution analysis of warping fields. The methodology allows the quantitative study of variation in characteristics previously only described from a qualitative perspective.
Table of Contents and Preface
(The Eurographics Association, 2005) Mike Chantler
Image classification using compression distance
(The Eurographics Association, 2005) Lan, Yuxuan; Harvey, Richard; Mike Chantler
The normalised compression distance measures the mutual compressibility of two signals. We show that this distance can be used for classification on real images. Furthermore, the same compressor can also operate on derived features with no further modification. We consider derived features consisting of trees indicating the containment and relative area of connected sets within the image. It had been previously postulated that such trees might be useful features, but they are too complicated for conventional classifiers. The new classifier operating on these trees produces results that are very similar to those obtained on the raw images thus allowing, for the first time, classification using the full trees.

Browse

Browsing VVG05 by Issue Date

Results Per Page

Sort Options