Vol. 15 (2006):
Abstracts of Papers
No. 3/4: Special Issue: Proc. ICCVG 2006 Conference.
Machine GRAPHICS & VISION, Vol. 15 (2006),
Megyesi Z., Kós G., Chetverikov D.:
3D reconstruction from images by normal aided matching.
MGV vol. 15, no. 1, 2006, pp. 3-28.
3D models play an increased role in today's computer applications. As a result,
there is a need for flexible and easy to use measuring devices that produce
3D models of real world objects. 3D scene reconstruction is a quickly evolving
field of computer vision, which aims at creating 3D models from images
of a scene. Although many problems of the reconstruction process have been
solved, the use of photographs as an information source involves some practical
difficulties. Therefore, accurate and dense 3D reconstruction remains
a challenging task. We discuss dense matching of surfaces in the case when
the images are taken from a wide baseline camera setup. Some recent studies
use a region-growing based dense matching framework, and improve accuracy
through estimating the apparent distortion by local affine transformations.
In this paper we present a way of using pre-calculated calibration data to improve
precision. We demonstrate that the new method produces a more accurate model.
Key words: image based reconstruction, 3D scene reconstruction,
stereo, wide-baseline, dense matching.
MCFI-based animation tweening algorithm for 2D parametric
motion flow/optical flow.
MGV vol. 15, no. 1, 2006, pp. 29-49.
In hand drawn animation, some of the animations are created at very low
frame rates (12 frames per second) because of a limited budget. A simple
technique for increasing frame rates is animation tweening. In fact,
most animation tweening applications process only vector-based objects.
Further, there is no algorithm which can automatically define extensive modelling
and animation of 3D objects for the animation tweening process. In this work,
a raster-based animation tweening algorithm which does not require extensive
modelling and animation of 3D objects is presented. The algorithm is based
on motion-compensated frame rate up-conversion (MC-FRC), whose performance
depends heavily on motion estimation (ME). A more accurate flow field,
which is a combination of 2D parametric motion flow field and optical flow field,
is developed. The results obtained for various types of hand drawn animation scenes
are illustrated. The experimental results show that higher quality in-between
frames can be obtained by the proposed algorithm.
Key words: computer animation, image processing, computer vision.
Practical camera calibration and image rectification in monocular
road traffic applications.
MGV vol. 15, no. 1, 2006, pp. 51-71.
In this paper we follow apractical approach to the problem of camera calibration
and image formation in a typical computer vision based traffic monitoring system.
Our study starts by analysing in detail the capturing scenario and the
undesirable effects resulting from the application of the pin-hole camera model
to traffic images. Based on the acquired experience, we present an integral
method to obtain the transformation matrices required to calibrate the camera
and to construct in real time a rectified and sub-sampled image where perspective
effects on the road plane have been removed. The motivation for this
perspective-free image is to reduce the amount of data which must be
computed (while avoiding loss of relevant information), remove influences from
objects external to the capturing area and simplify the operation of the
subsequent detection and tracking stages. The latter statement is justified
since vehicle shapes can now be approximated by rectangles, the distance between
any two neighbouring points on the road plane remains constant all over the image,
and parallel trajectories are restored in the rectified image.
Key words: camera calibration, image rectification,
road traffic monitoring, non-linear sub-sampling, perspective correction,
Hillion A., Roux C., Donescu I., Avaro O.:
Generalized second-order invariance in texture modeling.
MGV vol. 15, no. 1, 2006, pp. 73-97.
In image processing, micro-textures are generally represented as homogeneous
random fields, the term "homogeneous" indicating a second-order stationary random
process. However, such a formulation is restrictive, and does not allow
for the processing of anisotropic textures. The aim of this paper is to study
a generalization of second-order stationarity to second-order invariance under
a group of transforms, in order to apply this generalization to texture modeling
and analysis. The general formulation of second-order homogeneity
or G-invariance is given in relation to the framework of group theory.
Two approaches are derived, taking into consideration transitive groups
and generalized translations. For the latter approach, an important particular
case is outlined, in which a second-order G-invariant random field
X can be one-to-one associated to asecond-order stationary random field.
Some examples of interesting groups of transforms are given. Finally, Cholesky
factorization is applied for the synthesis of random fields showing
the generalized invariance property.
Key words: random field, second-order stationarity, group orbits,
texture modeling, texture synthesis.
He Z., Harada K.:
Solving point-feature labeling placement problem by parallel Hopfield
neural network on GPU graphics card.
MGV vol. 15, no. 1, 2006, pp. 99-120.
This paper discusses the application of parallel Hopfield neural networks
in solving the point-feature labeling placement (PFLP) problem by using programmable
graphics hardware found in a commodity PC. In this paper, we focus on two
aspects. The first aspect concerns mapping the PFLP onto parallel Hopfield neural
network. The second aspect is the detailed method of implementing the parallel
Hopfield neural network on graphics hardware. We demonstrate the effectiveness of
implementing the parallel Hopfield network by solving the PFLP problem. Moreover,
our proposal makes use of the advantages of the parallel Hopfield network on low-cost
Key words: parallel computing, Hopfield neural network, GPU, PFLP.
Machine GRAPHICS & VISION, Vol. 15 (2006),
Kwasnicka H., Paradowski M.:
Fast image auto-annotation with discretized
feature distance measures.
MGV vol. 15, no. 2, 2006, pp. 123-140.
A new model for the image
auto-annotation task is presented. The model can be classified as a
fast image auto-annotation one. The main idea behind the model is to
avoid various problems with feature space clustering. Both the image
segmentation and the auto-annotation process do not use any
clustering algorithms. The method presented here simulates
continuous feature space analysis with very dense discretization.
The paper presents the new approach and discusses the results
achieved with it.
Key words: image auto-annotation, fast auto-annotation,
image feature extraction.
Les Z., Les M.:
Shape understanding system: 3D interpretation as a part
of the visual concept formation.
MGV vol. 15, no. 2, 2006, pp. 141-175.
This paper presents a new method of interpretation of
the 2D visual objects in terms of 3D geometrical or real world
objects. The 3D interpretation of the visual objects depends on the
class where a given object is assigned. Each class has its own 3D
interpretation method. The interpretation methods for selected
classes are described and the results of testing of these methods
are presented. It is shown that a visual object can be interpreted
as a 3D object by assigning it to one of shape classes. The main
novelty of the presented method is that the process of
interpretation is related to a visual concept represented as a set
of symbolic names of the shape classes. The visual concept, which is
one of the components of the category of a visual object, makes it
possible to represent knowledge about the visual object in the form
of a categorical structure. The presented results are part of
research aimed at developing a shape understanding method that will
be able to perform complex visual tasks connected with visual
Key words: shape understanding, object recognition,
visual concept, visual reasoning.
Variants of pattern recognition and image understanding method
used in telescope guiding system.
MGV vol. 15, no. 2, 2006, pp. 177-195.
This paper presents a suggestion
for using image understanding (IU) methods in a telescope guiding
system, which is an example of a real-time system. In the presented
approach, the guiding system uses a knowledge base to describe a
time-varying situation and respond to it adequately. The large
domain of understanding (the space of image features and time) is
split to subspaces where the understanding tasks are simplified. The
use of the IU idea to control the telescope motors and to optimize
the system parameters is also discussed in the paper.
Key words: object recognition, object visual identification,
computer vision, learning in vision.
Miyazaki R., Harada K.:
Mesh construction based on differential surface property.
MGV vol. 15, no. 2, 2006, pp. 197-210.
In this paper, we propose a novel
method for extraction of a feature from range data to create of a
simplified triangular mesh. The feature extracted by our method is
based on the surface property. In the past, vertices on a highly
curved surface were extracted from range data, and the extracted
vertices were triangulated to construct a simplified triangular
mesh. However, the vertices extracted with such a method were not
enough to create a simplified triangular mesh in some cases. For
example, when the shape shows less change in one direction, like a
parabolic shape, too many vertices are extracted by the traditional
method. We introduce a surface property defined by a combination of
the mean curvature and the gaussian curvature. Vertices are
classified into three categories according to the surface property.
Then, the structures consisting of vertices and edges are extracted
suitably for constructing a simplified triangular mesh. Following
that, we can create the desirable triangulation through constrained
Delaunay triangulation using the extracted structures.
Key words: feature extraction, surface property,
mesh simplification, mesh segmentation.
Ramakrishnan S., Selvan S.:
Wavelet-based modeling of singular values
for image texture classification.
MGV vol. 15, no. 2, 2006, pp. 211-225.
A new algorithm based on the wavelet packet transform is proposed
for the classification of image textures. Energy matrices are formed
from subband coefficients of the wavelet packet transform. Singular
value decomposition is then employed on the energy matrices. The
probability density function of singular values is modeled as
exponential distribution, and the model parameter is estimated using
the maximum likelihood estimation technique. The model parameter,
one for each subband, is used to form the feature vector.
Classification is carried out using the Kullback-Leibler Distance
(KLD). Performance of the algorithm is compared with model-based
and feature-based methods in terms of the signal-to-noise ratio and
the classification rate. Experimental results prove that the
proposed algorithm achieves better classification rate under noisy
Key words: wavelet packet transform,
image texture classification, singular value decomposition.
Koprowski R., Wrobel Z.:
The automatic measurement of a staining reaction level.
MGV vol. 15, no. 2, 2006, pp. 227-238.
The paper presents an
application of an algorithm for automatic segmentation of
biological cell structures. The algorithm, based on two
morphological operations -- conditional opening and conditional
closing, is described in detail in (Koprowski R., Wrobel Z.:
Automatic segmentation of biological cell structures based on conditional
opening or closing. Machine Graphics and Vision, 14(3), 285-307)).
The results of the segmentation of biological cell structure images
as well as the evaluation of a staining reaction saturation level
and its metrological properties are studied.
New Books Notes
MGV vol. 15, no. 2, 2006, pp. 239-240.
[Robert Cierniak: Tomografia komputerowa: Budowa urzadzen CT, algorytmy rekonstrukcyjne
(Computed tomography: CT engineering, Reconstructive algorithms, in Polish).
Published by Akademicka Oficyna Wydawnicza EXIT, Warsaw 2005.
Machine GRAPHICS & VISION, Vol. 15 (2006),
Proc. International Conference on Computer Vision and Graphics (ICCVG 2006),
Warsaw, Poland, September 25-27, 2006.
Special issue editor: Konrad Wojciechowski.
- Wojciechowski K.:
MGV vol. 15, no. 3/4, 2006, pp. 243-244.
Al-Hamadi A., Niese R., Panning A., Michaelis B.:
Toward robust face analysis method of non-cooperative persons
in stereo color image sequences.
MGV vol. 15, no. 3/4, 2006, pp. 245-254.
[The winner of the Best Young Scientist's Presentation Award.]
This paper describes a
novel method for analyzing single faces of non-cooperative persons
on the basis of stereoscopic color images. The challenges arise from
the fact that the persons observed are non-cooperative, which in
turn complicates further processing as facial feature extraction and
tracking in image sequence. In our method, face detection is based
on color-driven clustering of 3D points derived from stereo. A mesh
model is registered with a post-processed face cluster, using a
variant of the Iterative Closest Point algorithm (ICP). The pose is
derived from correspondence. Then, the pose and model information
are used for face normalization and facial feature localization.
Automatic extraction of facial features is carried out using
modified Active Shape Models (ASM). In contrast to the simple ASM,
another approach is pursued in this work. It involves two
modifications to the ASM, which lead to greater stability and
robustness. The results show that stereo and color are powerful cues
for finding the face and its pose, and for facial feature extraction
under a wide range of poses, illumination types and expressions (PIE).
Key words: facial feature extraction, active shape models,
stereo and color image analysis.
Andrysiak T., Choras M.:
Algorithms for stereovision disparity calculation
in the moment space.
MGV vol. 15, no. 3/4, 2006, pp. 255-264.
This article presents various theoretical and experimental
approaches to the problem of stereo matching and disparity
estimation. We propose to calculate stereo disparity in the moment
space, but we also present numerical and correlation-based methods.
In order to calculate the disparity vector, we decided to use the
discrete orthogonal moments of Tchebichef, Zernike and Legendre. In
our research in stereo disparity estimation, all of those moments
were tested and compared. Experimental results confirm effectiveness
of the presented methods for determining stereo disparity and stereo
matching for robotics and machine vision applications.
Key words: stereovision, disparity calculation, moments space.
Arasteh S., Hung C.C.:
Color and texture image segmentation using uniform local
MGV vol. 15, no. 3/4, 2006, pp. 265-274.
The paper describes a new algorithm for image segmentation based
on the color and texture features. The Uniform Local Binary Pattern
(ULBP) method is used to extract texture features. Color features
are defined based on the pixels' color bands. Image segmentation is
carried out using the K-means algorithm on feature vectors,
including color and texture features. The distance measure is
defined as a function of the color and texture feature vector
distances from the K-means defined centers. The weighting parameter
is used to adjust the relative contribution of the color and texture
features. The proposed algorithm is applied to color images in the
RGB, HSV and IHLS color spaces. Experimental results show that the
proposed algorithm yields good performance in combining color and
texture features to distinguish different texture patterns. In
particular, for textures with high color contrast, the results are
prominent. The main advantage of the method is its speed and
simplicity, which are inherited from the K-means algorithm.
Key words: color and texture segmentation,
local binary patterns.
Bando T., Oshima T.:
Reproduction of global impression of landscape images
by improved texture reconstruction methods.
MGV vol. 15, no. 3/4, 2006, pp. 275-284.
In order to reproduce the atmosphere or the global impression of
the landscape images we have created model patterns by
reconstructing the texture of the whole images. We have first
segmented full-color landscape images into patches of same color
after converting them into 216-color images, and then analyzed the
size and the center of gravity of each color patch. We have created
five improved reconstructed model patterns from the data in the
color patches. Some of the reconstructed texture patterns are
similar in the global appearance to the original landscape images,
although the details of the original images have been completely
destroyed, to the extent that it is difficult to understand what
these images represent at all. Each of the reconstructed patterns
has been evaluated in order to find good methods for reproducing
the atmosphere of the landscape. The patterns reconstructed using
the proposed methods, which take into consideration the shape and
position of the original color patches, demonstrate very high
quality. The results suggest that the reconstructed global texture
of the whole image creates an atmosphere or a global impression of
the landscape images.
Key words: landscape, texture reconstruction, atmosphere.
Bojar K., Nieniewski M.:
Modelling the spectrum of the fourier transform of the texture
in the solar EIT images.
MGV vol. 15, no. 3/4, 2006, pp. 285-295.
We present a
phenomenological parametric model for the spectrum of the discrete
Fourier transform (DFT) of the images obtained from the Extreme UV
Imaging Telescope (EIT) of the SOHO mission. As this spectrum
decays very fast, we model its logarithm rather than the original
spectrum. The proposed model is rotation-invariant. The vicinity of
the direct current (DC) component of the logarithm of the spectrum
is modelled by the sum of two exponential functions, while the
region of high frequencies is modelled by a single exponential
function summed with a constant. We discuss a method for fitting the
model to the experimental data, show the results of numerical
experiments, and discuss various measures of goodness of the fit.
The fitting of the described model was carried out for a sequence of
images covering one year, and the time evolution of the measures of
goodness of fit is also presented.
Key words: texture, DFT, modelling, EIT solar images.
Boyer V., Sobczyk D.:
A Model to Blend Renderings.
MGV vol. 15, no. 3/4, 2006, pp. 297-304.
We propose a model to blend renderings. It consists in mixing
different kind of rendering techniques in the same frame to enhance
the visualisation of information on large scenes. This method can be
implemented in real-time during the rendering process using the GPU
programming. Moreover, the rendering techniques used and the key
points defined by the user can be interactively changed. In this
paper we present the model, a new non-photorealistic rendering
technique and images produced by our method.
Key words: non-photorealistic rendering GPU.
Robust Filtering in a Laser Scanner Point Cloud.
MGV vol. 15, no. 3/4, 2006, pp. 305-310.
This paper presents a new
effective and robust approach to noise reduction in a
three-dimensional data measurement algorithm. In the literature,
there are numerous algorithms for noise reduction. The proposed
filter class is based on the nonparametric estimation of the density
probability function in a sliding filter sphere. The main idea of
the applied algorithm depends is to maximize the distance between
points in the three-dimensional space - the nearest neighbors in
sliding 3D sphere.
Key words: 3D pre-processing algorithms,
3D acquisition, cloud of points filtering.
Cinque L., Sangineto E., Tanimoto S.:
Recognition of articulated robots in the RoboCup domain.
MGV vol. 15, no. 3/4, 2006, pp. 311-320.
This paper presents a system for articulated object recognition.
The system has been tested in the RoboCup domain (four-legged
league), an international competition among autonomous dog-like
robots playing soccer. Nevertheless, the proposed method does not
depend on any specific domain, but is thought to be applicable to
generic objects composed of rigid subparts linked by rotational
and/or translational joints.
Key words: articulated object recognition, RoboCup,
deformable shapes, model-based approach.
Cuesta-Frau D., Hernández-Fenollosa M. A., Vicedo-Payá J.,
Particle measurement in scanning electron microscopy images.
MGV vol. 15, no. 3/4, 2006, pp. 321-328.
This paper describes the segmentation of nanoparticles of ZnO
obtained by mechanical milling. Segmentation of objects in images is
a common application of computer vision methods. In contrast to
manual segmentation, these techniques are fast, objective, and
accurate. We describe in this paper a method based on such
techniques aimed at segmenting the particles in a microscopic image
of ZnO in order to obtain an approximation of the grain size, and a
measure of the homogeneity, in a non-supervised way. The images are
obtained using scanning electron microscopy and then preprocessed to
enhance the contrast and to reduce the noise. Next, an edge
detection algorithm is applied to obtain the boundaries of the
particles. Finally, the particles that satisfy a specific criterion
are extracted and measured, and their measure is taken as an
approximation of the particle size.
Key words: image processing, edge detection, pattern matching.
Hardware-software system for acceleration
of image processing operations.
MGV vol. 15, no. 3/4, 2006, pp. 329-337.
The paper presents design
and architecture of a hybrid software/hardware system for
acceleration of image processing. The front end consists of a
software interface that defines the basic data structures and
exchange mechanisms for connecting to external software. The back
end consists of a hardware board which is responsible for
acceleration of image computations. Thus, the two main components
follow the handle/body concept, which allows modifications to the
implementation without changes in interfaces. This flexibility
allows for better resource usage, and faster development, and
facilitates system extensions. In this paper we present the design
and implementation issues for the system, as well as discuss its
run-time performance for the selected image operations.
Key words: image processing library, hardware acceleration.
Detection and segmentation of moving vehicles and trains using
gaussian mixtures, shadow detection and morphological processing.
MGV vol. 15, no. 3/4, 2006, pp. 339-348.
The solution presented in
this paper combines background modelling, shadow detection and
morphological and temporal processing into a single system
responsible for detection and segmentation of moving objects
recorded with a static camera. Vehicles and trains are detected
based on their pixel-level difference with respect to a continually
updated background model, using a Gaussian mixture calculated
separately for every pixel. The shadow detection method utilizes a
colour model which allows for estimating chromatic and brightness
differences between the pixel colour and the background model.
Morphological and temporal operations performed on binary images
denoting moving objects include connecting the components, closing
and temporal filtering. Experiments carried out involve employing
implemented algorithms to detect vehicles and trains in video
sequences. The results achieved are described and illustrated in
Key words: image processing, vehicle detection and segmentation,
mixture of Gaussians, shadow detection, morphological operations.
El-Etriby Sh., Al-Hamadi A., Michaelis B.:
Dense depth map reconstruction by phase difference-based algorithm
under influence of perspective distortion.
MGV vol. 15, no. 3/4, 2006, pp. 349-361.
This paper describes a novel phase difference-based algorithm
applied to the corresponding points in two views, which takes into
account the surface perspective distortion (foreshortening).
The challenges arise from the fact that stereo images are acquired
from slightly different views. Therefore, the surface of a
projected image is more compressed and occupies a smaller area in
one view. Since the projective distortion region can not be
estimated in terms of a fixed-size matching algorithm, we suggest
using a local spatial frequency representation model to address
this problem. Instead of matching intensities directly, a Gabor
scale-space expansion (scalogram) is used. The scalogram expresses
the filter output as a function of the spatial position and the
principal wavelength to which the filter is tuned. The phase
difference at the corresponding points in the two images is used to
find the disparity value. The suggested algorithm provides an
analytical closed-form expression for the perspective foreshortening
effect. The foreshortening factor is verified to overcome the
perspective distortion region. The efficiency and performance of the
suggested algorithm for dense depth map reconstruction is
demonstrated on the basis of analysis of rectified real images.
Hence, our proposed method has a superior performance in comparison
with other conventional methods.
Key words: stereo matching, disparity estimation,
3D reconstruction, perspective distortion.
Fedoseev V., Chernov V.:
Cryptography and canonical number systems in quadratic fields.
MGV vol. 15, no. 3/4, 2006, pp. 363-372.
This paper proposes an encryption method based on
representation of messages in the canonical number systems (CNS) in
quadratic fields. The essence of the encryption method is conversion
of the representation of integers from the conventional number
system to their representation in the CNS in a certain quadratic
field. A sufficiently wide range of CNS with a given number of
digits ensures resistance of the method to "accidental guessing"
of the secret keys. Nonlinear nature of the conversion ensures its
resistance to frequency analysis.
Key words: canonic number systems, cryptography,
Florek A., Piascik T.:
Efficient object description and recognition based on shape signature.
MGV vol. 15, no. 3/4, 2006, pp. 373-380.
A simple and efficient
approach to recognition and classification of planar shapes is
proposed. This approach is based on comparison of areas of dynamic
adapted centroid signatures. The usefulness of our approach for
assessment of the similarity equally of convex and non-convex
objects and objects containing openings is shown.
Key words: shape representation, signature, object recognition.
Grundland M., Vohra R., Williams G.P., Dodgson N.A.:
Nonlinear multiresolution image blending.
MGV vol. 15, no. 3/4, 2006, pp. 381-390.
We study contrast enhancement for multiresolution image
blending. In image compositing, image stitching, and image fusion, a
blending operator combines coefficients of a pixel array, an image
pyramid, a wavelet decomposition, or a gradient domain
representation. Linear interpolation reduces variation and thereby
causes contrast loss, while coefficient selection increases
variation and thereby causes color distortion. Offering a continuous
range of possibilities between these standard alternatives, the
signed weighted power mean enables the user to calibrate the
contrast of composite images.
Key words: multiresolution contrast enhancement,
image compositing, image blending, image fusion, image stitching,
image pyramid, image editing, video editing, cross dissolve, digital art.
- Hidot S., Lafaye J.-Y., Saint-Jean C.:
Discriminant factor analysis for movement recognition:
application to dance.
MGV vol. 15, no. 3/4, 2006, pp. 391-399.
In this paper, we present an example of operator discriminant
factor analysis applied to the study of movement. Our goal is to
contribute to the automatic recognition of typical movements and
give analytic elements to discuss the differences between the
various types. As a signature for the movements, we select the
'Relational Covariance Operator' and explain our motivation.
Discriminant analysis cannot be directly applied to covariance
operators; so, in order to deal with a well conditioned problem and
avoid useless heavy computation, we proceed in two steps. We first
achieve a variant of principal components analysis on covariance
operators by using convenient metrics, and then apply plain linear
discriminant analysis on the resulting axis. Such a method allows a
simple and powerful interpretation of the movement typology. We also
present an experiment on motion-captured movements performed by
dancers from the 'Ballet Atlantique Régine Chopinot'.
Key words: discriminant analysis, operator analysis,
movement signature, movement recognition.
Hnat K., Porquet D., Merillou S., Ghazanfarpour D.:
Real-time wetting of porous media.
MGV vol. 15, no. 3/4, 2006, pp. 401-413.
Studying light reflection properties is a crucial factor in
achieving a high degree of realism in image synthesis. Considered as
a challenge in itself, it becomes even more complicated when dealing
with specific changes in appearance due to external factors. Among
these changes, one of the most common is wetting: surfaces appear
darker and more specular as their wetting level increase. Such a
phenomenon is of great visual importance in outdoor scenes under
rain falls, for example. This change in appearance is mainly due to
the porous nature of surfaces. In this paper, we propose to handle a
porous surface BRDF post-process model in real-time and to extend it
to account for wetting, with simple and intuitive parameters.
Key words: shading, porosity, weathering, wetting,
Ivanov Y., Hamid R.:
Weighted ensemble boosting for robust activity recognition
MGV vol. 15, no. 3/4, 2006, pp. 415-427.
In this paper we
introduce a novel approach to classifier combination, which we term
Weighted Ensemble Boosting. We apply the proposed algorithm
to the problem of activity recognition in video, and compare its
performance to different classifier combination methods. These
include Approximate Bayesian Combination, Boosting, Feature
Stacking, and the more traditional Sum and Product rules. Our
proposed Weighted Ensemble Boosting algorithm combines the Bayesian
averaging strategy with the boosting framework, finding useful
conjunctive feature combinations and achieving a lower error rate
than the traditional boosting algorithm. The method demonstrates a
comparable level of stability with respect to the classifier
selection pool. We show the performance of our technique for a set
of 6 types of classifiers in an office setting, detecting 7
classes of typical office activities.
Key words: video analsyis, activity recognition,
boosting, ensemble methods.
Iwanowski M., Huk A.:
Satellite image pansharpening by chrominance propagation
combined with kernel interpolation.
MGV vol. 15, no. 3/4, 2006, pp. 429-438.
This paper describes a method for propagation-based spatial
interpolation of missing color information in satellite images. Most
of the land-observation satellites produce two types of imagery for
every scene -- multispectral and panchromatic. The first kind is
characterized by lower spatial resolution but higher spectral one,
while the second is a graylevel image at higher resolution. In order
to get full color visualization of a scene, the high-resolution
panchromatic image must be combined with the low-resolution color
information taken from multispectral bands. This process is called
pansharpening. In this paper, a new method for pansharpening
is proposed which combines chrominance propagation with kernel
interpolation. Thanks to the propagation step, the method properly
reconstructs color information and does not blur the edges on color
Key words: remote sensing, pansharpening, interpolation.
Jiménez-López F., Cuesta-Frau D., Linares-Pellicer J., Micó-Tormos P.:
A comparative study of local thresholding methods for document
MGV vol. 15, no. 3/4, 2006, pp. 439-450.
This paper provides a
comparative study of several local thresholding techniques
applied in the binarization of document images with text and graphs.
First, the methods are described, and then a quantitative comparative
analysis is carried out. Ground truth images are used in the
experiments to provide an objective measure, and computational costs are
Key words: image processing, image binarization.
Kasprzak W., Szynkiewicz W., Czajka L.:
Rubik's cube reconstruction from single view for service robots.
MGV vol. 15, no. 3/4, 2006, pp. 451-459.
The Rubik's cube puzzle
is seen as a benchmark for service robots. In such an application, a
computer vision subsystem is required to locate the object in space
and to determine the configuration of its colored cells. This paper
presents a robust algorithm for Rubik's cube reconstruction from a
single view in real time. An issue of special interest is to obtain
a good tade-off between the quality of results, and the
computational complexity of the algorithm.
Key words: color analysis, shape description, segment grouping,
service robots, surface orientation, 3-D object recognition.
Korbel P., Slot K.:
Cellular neural network-based object recognition with deformable grids.
MGV vol. 15, no. 3/4, 2006, pp. 461-469.
The following paper presents an idea for a parallel implementation
of the deformable grid paradigm within the framework of Cellular
Neural Networks. Parallel processing may alleviate the problem of
high complexity of deformable template matching and significantly
speed up object recognition tasks. The paper presents details of a
CNN-based implementation of the basic element of the deformable
grid-based image processing, which is image-grid matching. Estimated
execution speed of the CNN-based method and recognition rates
achieved in the experiments make the method an attractive framework
for applications such as high-speed coarse object classification.
Key words: object recognition, image processing,
cellular neural networks.
Ksantini R., Ziou D., Colin B., Dubeau F.:
Weighted pseudo-metric for a fast CBIR method.
MGV vol. 15, no. 3/4, 2006, pp. 471-480.
In this paper, a simple and fast querying method for content-based
image retrieval is presented. In order to measure the similarity
degree between two color images both quickly and effectively, we use
a weighted pseudo-metric employing one-dimensional Daubechies
decomposition and compression of the extracted feature vectors. In
order to improve the discriminatory capacity of the pseudo-metric,
we compute its weights using separately a classical logistic
regression model and a Bayesian logistic regression model. The
Bayesian logistic regression model was shown to be significantly
better than the classical logistic regression model at improving
the retrieval performance. Experimental results are reported on the
WANG and ZuBuD color image databases proposed by
[Deselaers T., Keysers D., Ney H.: Classification error rate for quantitative
evaluation of content-based image retrieval systems. 17th International
Conference on Pattern Recognition (ICPR'04), 2, pp. 505-508,
Key words: color image retrieval, weighted pseudo-metric,
logistic regression models.
Léon P.-F., Skapin X., Meseure P.:
Topologically-based animation for describing geological evolution.
MGV vol. 15, no. 3/4, 2006, pp. 481-491.
This paper presents a
topologically-based animation system which aims at representing the
topological evolution of structured objects over time. Its robustness
and versatility both relie on the $n$-dimensional generalised map
formalism. The animation is modelled as a series of maps ordered in
time and represents each topological modification of the
structure. We define several dedicated topological operations, which are
translated into a script defining the animation. We finally show the
usefulness of the approach by means of a specific application in
geology, namely representation of a subsoil evolution in 2D.
Key words: topologically-based animation,
generalised maps, geology.
Minimizing CPU Usage in Soft Shadow Volumes Algorithm.
MGV vol. 15, no. 3/4, 2006, pp.493-503.
In this paper we present methods
for reducing the CPU burden associated with the original
implementation of the hardware-accelerated soft shadow volumes
algorithm. We propose and implement two hardware-accelerated wedge
construction methods: using the programmable vertex and fragment
pipelines. New algorithms reducing the number of draw calls needed
to render the shadow region (to three, two or even one) are also
described. Tests performed using the current consumer level hardware
show that our algorithms are no longer CPU-bound, and therefore can
be over 2 times faster than the original algorithm.
Key words: shading, soft shadows, shadow volumes, graphics hardware.
Marusiak K., Szczepanski M.:
Video based road traffic detection and analysis.
MGV vol. 15, no. 3/4, 2006, pp. 505-514.
In this article an autonomous vehicle detection algorithm is
presented. The described technique is capable of accurate vehicle
detection, can work with different cameras and is fairly immune to
illumination changes. It is based on the idea of an adaptive
background model and the background subtraction method for detecting
motion. The application's performance tests ware conducted using
real life video sequences and gave satisfactory results.
Key words: video processing, motion detection,
traffic monitoring, background adaptation, background detection, directShow.
- Mischler D., Romaniuk B., Benassarou A., Bittar E.:
Robust 4D segmentation of cells in confocal images.
MGV vol. 15, no. 3/4, 2006, pp. 515-524.
We present a method to automatically extract the evolution of the
cell envelope in 4D confocal images. Our method is based on
4DReAM, a tracking system consisting of a previously presented
deformable surface model, which can change its topology. The process
consists in attracting the model for each volume of a 4D series
towards an iso-surface of interest, and towards the image gradients.
We then statistically estimate the characteristics of the cell
surface on the nodes of the model, and reconstruct it. We show
detailed results on the segmentation of the cell envelope during the
Key words: deformable model, 4D tracking,
living cells, automatic segmentation, gradient comparison, differential operators.
Mokrzycki W.S., Salamonczyk A.:
Generating 3D multiview exact polyhedron representation
by scanning faces surroundings.
MGV vol. 15, no. 3/4, 2006, pp. 525-536.
This article presents 3D multiview exact models that are complete
representations of a polyhedron obtained by the means of the faces
surroundings scanning algorithm. The algorithm uses the viewing
sphere and the perspective projection concepts (known as the
K-M view space model). Such models will be used for visual
identification based on them and a scene depth map. We develop the
concept and give an algorithm for face-dependent generation of 3D
exact views polyhedron representation.
Key words: visual object identification, depth map,
precise 3D multiview polyhedron models,
viewing sphere with central projection,
models completion state of multiview representation.
Mousa M., Chaine R., Akkouche S.:
Frequency-based representation of 3D point-based surfaces
using spherical harmonics.
MGV vol. 15, no. 3/4, 2006, pp. 537-546.
In this paper, we propose a precise frequency-based representation
for oriented point-based surfaces using spherical harmonics. The
representation can be useful in many applications, such as
filtering, progressive transmission and coding of 3D surfaces. The
basic computation in our approach is the spherical harmonics
transform of local spherical radial functions induced by a set of
points. An important feature of our approach is that the
calculations are performed directly on local 2D triangulations of
the point-based surface without any prior space voxelization. This
property ensures that the complexity of our computation of the
spherical harmonics transform is linear in the number of triangles
in the local patch. We present some experimental results which
demonstrate our technique.
Key words: spherical harmonics, point-based surface,
direct simplex-based computation, surface reconstruction, geometric texture.
Nacereddine N., Hamami L., Ziou D., Tridi M.:
Probabilistic deformable models for weld defect contour estimation
MGV vol. 15, no. 3/4, 2006, pp. 547-556.
This paper describes a novel method for segmentation
of weld defect in radiographic images. Contour estimation is
formulated as a statistical estimation problem, where both the
contour and the observation model parameters are unknown. Our
approach can be described as a region-based maximum likelihood
formulation of parametric deformable contours. This formulation
provides robustness against the poor image quality, and allows
simultaneous estimation of the contour parameters together with
other parameters of the model. Implementation is performed by a
deterministic iterative algorithm with minimal user intervention.
Results testify very good performance of such contour estimation
Key words: Gaussian and Rayleigh distributions,
contour estimation, maximum likelihood, parametric deformable contours.
- Nacereddine N., Hamami L., Ziou D.:
Thresholding techniques and their performance evaluation
for weld defect detection in radiographic testing.
MGV vol. 15, no. 3/4, 2006, pp. 557-566.
In non-destructive testing with radiography, a perfect
knowledge of the weld defect shape is an essential step to
appreciate the quality of the weld and make decision on its
acceptance or rejection. Because of the complex nature of the
considered images, and in order that the detected defect region
represent the real defect as accurately as possible, the choice of
the thresholding methods must be made judiciously. In this paper,
performance criteria are used to conduct a comparative study of the
thresholding methods based on the gray level histogram, the 2D
histogram and the locally adaptive approach to weld defect detection
in radiographic images.
Key words: 1D and 2D histogram, locally adaptive approach,
performance criteria, radiographic image, thresholding, weld defect.
Lip-reading with discriminative deformable models.
MGV vol. 15, no. 3/4, 2006, pp. 567-575.
The following paper describes a novel lip-reading method
developed for the purpose of isolated word recognition. The method
is based on a concept of a discriminative deformable model, which
represents an image analysis method derived from the deformable grid
paradigm. The discriminative deformable model is used to
characterize the lip shape at each frame of the video sequence. The
information extracted from the consecutive frames is next analyzed
using the Hidden Markov Models. The proposed visual speech
recognition method is tested using the Polish digits recognition
Key words: lip-reading, deformable grid, speech recognition.
Pelc L., Kwolek B.:
Recognition of actions in meeting videos using timed automata.
MGV vol. 15, no. 3/4, 2006, pp. 577-584.
This paper addresses the
problem of action recognition in meeting videos. A declarative
knowledge provided graphically by the user together with person
positions extracted by a tracking algorithm are used to generate the
data for recognition. The actions have been formally specified using
timed automata. The specification was verified on the basis of
simulation tests as well as an analysis. The tracking is
accomplished using a particle filter built on cues such as color,
gradient and shape.
Key words: action recognition, vision-based people tracking,
An effective edge-directed frequency filter for removal of aliasing
in upsampled images.
MGV vol. 15, no. 3/4, 2006, pp. 585-598.
Raster images can have a range of various distortions
related to their raster structure.
Their upsampling might in effect substantially reveal
the raster structure of the original image, known as
aliasing. The upsampling itself may introduce aliasing
into the upsampled image as well.
The presented method attempts to remove the
aliasing using frequency filters based on discrete fast Fourier
transform, and applied directionally in certain regions placed along
the edges in the image.
As opposed to some anisotropic smoothing methods, the presented
algorithm aims to selectively reduce the aliasing only, preserving
the sharpness of image details.
The method can be used as a post-processing filter along with various
upsampling algorithms. It was experimentally shown that the method can
improve the visual quality of the upsampled images.
Key words: aliasing, upsampling, frequency filter, edge detection.
Frame rate up-conversion using region-based optical flow.
MGV vol. 15, no. 3/4, 2006, pp. 599-606.
Frame rate up-conversion is widely used for converting between the
standards and increasing the frame rate of low bit rate compression
video conferences. The raster-based animation tweening algorithm
can not be directly applied with frame rate
up-conversion, because the crucial procedure of the algorithm, the
mean shift-based image segmentation method, can not handle complex
characteristics of a real world video image. In some case, the
background and foreground may be merged into one segment. In this
work, a synergistic image segmentation method is used according to
the particular needs of a real world video image. The experimental
results illustrate that higher quality in-between frames can be
obtained by the proposed algorithm.
Key words: image processing, computer vision.
Stachera J., Rokita P.:
Fractal-based hierarchical mip-pyramid texture compression.
MGV vol. 15, no. 3/4, 2006, pp. 607-619.
As the level of realism of computer-generated images increases
with the number and resolution of textures, we faced the problem of
limited hardware resources. Additionally, filtering methods require
multiple access and extra memory space for texture representation,
thus severely reducing the memory space and bandwidth, the most
common example being mip-mapping technique. We propose a
hierarchical texture compression algorithm for real-time
decompression on the GPU. Our algorithm is characterised by low
computational complexity, random access and a hierarchical
structure, which allows access to the first three levels of an
encoded mip-map pyramid. The hierarchical texture compression
algorithm HiTC is based on a block-wise approach, where each block
is subject to local fractal transform and further effectively coded
by one level of the Laplacian Pyramid.
Key words: texture compression, fractal compression,
Laplacian pyramid, mip-mapping.
Mesh representation of simply connected 3D objects in visualisation
of melting phenomena.
MGV vol. 15, no. 3/4, 2006, pp. 621-630.
This paper discusses
some aspects of the physics approach to simulation of the melting
phenomena: energy distribution in a volumetric model, substance
vaporization and their relation to geometric properties of a model.
A new simulation method for melting objects is presented. The method
consists in mesh deformations which allow simulation of melting
phenomena with preservation of visual effects similar to those
achieved in the volumetric approach. Moreover, a few examples of
melted models are included to show visual effects achievable with
the help of the method. Meshes used with the method should be simply
connected sets. This means there are no holes in them, and that they
constitute a single object through the whole simulation process.
Furthermore, a new method for determining whether a point is
situated inside a given polygon is shown.
Key words: mesh, melting, polygon inclusion.
Genetic filters for video noise reduction.
MGV vol. 15, no. 3/4, 2006, pp. 631-638.
This paper describes a genetic programming approach to video
filtering, based on evolution programming for image sequence
filter design. Each gene represents a filter weight. The intra- and
interframe approaches are shown and discussed.
Key words: genetic algorithms, image enhancement, nonlinear filters.
Corner-based keypoints for scale-invariant detection of partially
MGV vol. 15, no. 3/4, 2006, pp. 639-647.
Local features (also known as
interest points, keypoints, etc.) are a popular and powerful tool
for matching images and detecting partially occluded objects. While
the problems of photometric distortions of images and rotational
invariance of the features have satisfactory solutions,
satisfactorily simple scale-invariant algorithms do not exist yet.
Generally, either computationally complex methods of scale-space
(multi-scale approach) are used, or the correct scale is estimated
using additional mechanisms. The paper proposes a new category of
keypoints that can be used to develop a simple scale-invariant
method for detecting known objects in analyzed images. Keypoints are
defined as locations at which selected moment-based parameters are
consistent over a wide range of different-size circular patches
around the keypoint. While the database of known objects (i.e. the
keypoints and their descriptions) is still built using a
multi-scale approach, analyzed images are scanned using only a
single-scale window and its sub-window. The paper focuses on the
keypoint building and keypoint matching principles. Higher-level
issues of hypotheses building and verification (regarding the
presence of objects in analyzed images) are only briefly discussed.
Key words: local features, keypoints, moment invariants,
scale-invariance, geometric approximation.
Tworzydlo J., Skabek K., Luchowski L., Winiarczyk R.:
Bringing into register incomplete range images of cultural artifacts.
MGV vol. 15, no. 3/4, 2006, pp. 649-658.
Finding solutions for
preserving cultural heritage and historic and architectural sites as
3D models is an important problem in our days. Building 3D models of
real objects is also an important research issue in virtual reality.
Since a 3D range scanner, used to obtain 3D data, is a line-of-sight
instrument, in most cases it is necessary to scan the object from
multiple viewpoints to completely reconstruct it. The acquired range
images need to be brought into register to build one coherent model.
This article presents two registration methods for incomplete range
images. One is based on the assumption that the range images being
registered have a significant overlap with some feature points. The
second one needs additional 2D photographs of the scene to perform
the right registration, but the overlapping area of the range images
is not necessary here. The presented methods are semi-automatic and
base on the analysis of redundant and uncertain data.
Key words: 3D scanner, registration of range images.
Wcislo R., Zelechowski M.:
Hardware based real-time non-rigid body animation.
MGV vol. 15, no. 3/4, 2006, pp. 659-664.
We describe a method for animation of elastic objects based on a
mass-spring system with a pressure model. The method employs two
different implementation approaches: CPU- and GPU-based.
Three-dimensional, macroscopic objects with a volume of arbitrary
geometry are considered. The main goal of the presented work is to
make use of a modern graphics card equipped with a programmable GPU
(Graphical Processor Unit).
Key words: non-rigid, real-time, animation, hardware based.
Potential field based camera collisions detection
in a static 3D environment.
MGV vol. 15, no. 3/4, 2006, pp. 665-672.
Collision detection methods usually need a long pre-calculation
stage or require difficult and time-consuming real-time computation.
Moreover, their effectiveness decreases as the complexity of the
scene increases. The seemingly promising solutions presented in
literature which are based on potential fields do not offer a
satisfactory functionality. This paper introduces a method which
provides a new potential field construction affecting camera
movement, which lets the viewer reach objects without constraints
and protects the user from getting into their structure.
Additionally, the proposed method comprises an easily executable
pre-calculation stage, and represents a solution independent of the
Key words: collision detection, potential field, navigation.
Zhang S., Chen Y.:
Image denoising based on wavelet support vector regression.
MGV vol. 15, no. 3/4, 2006, pp. 673-680.
Denoising is an important application of image processing. We have
constructed a denoising system which learns an optimal mapping from
the input data to denoised data. The Morlet wavelet was used as the
kernel function to construct the wavelet support vector machine. The
noised image data is mapped to denoised values by wavelet support
vector regression. The result shows that denoising via wavelet
support vector regression could perform better than Gaussian
smoothing, median filtering and average filtering on the
experimental image and it also performs better than Gaussian radial
basic function support vector regression.
Key words: image denoising, support vector regression,
wavelet analysis, function approximation.
Zabinski T., Grygiel T., Kwolek B.:
Design and implementation of visual feedback for an active tracking.
MGV vol. 15, no. 3/4, 2006, pp. 681-690.
Active visual tracking is used to direct the attention of the camera
to an object and maintain it in the camera's field of view. A
steered camera is used to decrease relative motion of the target in
the image plane. This leads to better performance of the mean-shift
based tracking algorithm, which requires the object tracked in the
current and the previous frame to overlap. A classical PID
controller and a nonlinear fuzzy controller have been tested in
steering the camera head.
Key words: active vision, vision-based tracking, visual servoing.
- Reviewers' index
- Authors' index
- Contents of volume 15, 2006
Maintained by Zenon Kulpa
Last updated Jan 11, 2008