Abstracts of Papers

No. 1,

No. 2,

No. 3: Special Issue on
**Stereogrammetry and Related Topics**,

No. 4.

9 (2000) | main | 11 (2002) |

Reprints of the papers may be obtained from their authors; contact

**Editorial Office, MGV**

Institute of Computer Science

ul. Ordona 21

01-237 Warszawa, Poland`<mgv@ipipan.waw.pl>`

- Haider Md.A., Kaneko T.:

**Reconstruction of 3D human hair shape from video captured images and CT data**.

MGV vol. 10, no. 1, 2001, pp. 3-14. -
We propose a three-dimensional (3D) reconstruction method of human
hair-shape from rotating head multiple video captured images and CT data.
It is well known that no hair is present on the polygonal skin surface
of the human head (
*3D-head*) reconstructed from CT or MRI data. Our task is to reconstruct and add the hair-shape on the 3D-head to create a realistic human head model for simulating post-surgical facial expressions. Using a sculpturing technique based upon rotating head images we propose a method of reconstructing the hair-shape as well as the concave and semi-occluded parts in the skin-hair junction regions. We have utilized binarized voxel data of the 3D-head (*solid-head*) in this regard. The sculpturing object in our definition is the solid-head surrounded by assumed thick hair-voxels. We sculpture the surrounding hair-voxels according to the extracted hair-region from the video captured images while keeping the internal solid-head intact. We reconstruct the concave and semi-occluded regions by digging up to the visible skin surface of the solid-head in/near the hair region. We define*complete-model*as the 3D polygonal surface obtained from solid-head including the residue sculptured hair-voxels on it. Experimentally we have shown that our method yields hair shape to be used in practice.

**Key words**: 3D reconstruction, 3D head modeling, 3D hair modeling, visualization. - Tefera D.A., Harada K.:

**A complement to the 8 point algorithm in matching uncalibrated image sequences of a scene**.

MGV vol. 10, no. 1, 2001, pp. 15-27. -
This paper presents an algorithm which generates an initial set
of matching corner points between two uncalibrated images
of the same scene. Matching different images of a single scene remains
one of the bottlenecks in computer vision. The major part of matching
algorithms is to recover the epipolar geometry, which heavily depends
on the accuracy of the set of eight corresponding points that are used
to calculate the fundamental matrix between the sets of images under
consideration. Thus, an algorithm to locate a reliable initial
*guess*for matching corner points is proposed. The algorithm has been tested and shown reliable with different types of images indicating its ability to localize reliable sets of matching corner points. With the help of non-maximum response suppression technique our approach yields better results than other methods.

**Key words**: matching, correspondence problem, epipolar geometry, corners, bubble sorting. - El-henawy I., Fouda Y.M., Enab Y.M.:

**A neurocomputing approach to the correspondence problem in stereo vision based upon an unsupervised neural network**.

MGV vol. 10, no. 1, 2001, pp. 29-46. -
The stereo matching problem is one of the most widely studied problems
in stereo vision. In this paper we introduce a neurocomputing approach
to the local stereo matching problem using edge segments as features
with several attributes. Most classical local stereo matching techniques
use features representing objects in both images and compute the minimum
values of attribute differences. Pajares et al. had verified that
the differences in attributes, for the true matches, cluster in a cloud
around a center. We use the self-organizing neural network to get
the best cluster center. Based on the similarity constraint, we compute
the minimum Mahalanobis distances between the differences of the attributes
for a new pair of features and the cluster center to classify
this new pair as true or false match.
Experimental results with two real pairs of images are shown.

**Key words**: computer vision, stereo vision, matching, unsupervised neural network. - El-Bakry H.M.:

**Automatic human face recognition using modular neural networks**.

MGV vol. 10, no. 1, 2001, pp. 47-73. -
In this paper, a fast biometric system for personal identification
through face recognition is introduced. In the detection phase,
a fast algorithm for face detection is combined with cooperative modular
neural networks (MNNs) to enhance the performance of the detection process.
A simple design for cooperative modular neural networks is described
to solve this problem by dividing the data into three groups. Furthermore,
a new faster face detection approach is presented through image decomposition
into many sub-images and applying cross correlation in frequency
domain between each sub-image and the weights of the hidden layer.
For the recognition phase, a new concept for rotation invariant based
on Fourier descriptors and neural networks is presented. Although
the magnitude of the Fourier descriptors is translation invariant,
there is no need for scaling or translation invariance. This is because
the face sub-image (20x20 pixels) is segmented from the whole image
during the detection process. The feature extraction algorithm based
on Fourier descriptors is modified to reduce the number of neurons
in the hidden layer. The second stage extracts wavelet coefficients
of the resulted Fourier descriptors before application to neural network.
The final vector is fed to a neural net for face classification.
Moreover, a modified hierarchy soft decision tree of neural networks
is introduced for face recognition. Compared with previous results,
the proposed algorithm shows good performance on recognizing human
faces with glass, beard, rotation, scaling, occlusion, noise, or change
in illumination. Also, the response time is reduced.

**Key words**: face detection, modular neural networks, image decomposition, local image normalization, face recognition, rotation invariance. - Kovács T.:

**Accuracy prediction in a 3D active triangulation scanner**.

MGV vol. 10, no. 1, 2001, pp. 75-87. -
An active triangulation 3D scanner was developed at the Budapest
University of Technology and Economics, Department of Automation.
In the first part of this paper the configuration of the 3D scanner
system will be presented. Then the edge tracking part of model
extraction algorithm will be described, especially focusing
on the handling of noise. Accuracy of the generated CAD model depends
on configuration parameters and the noise of the frame buffer image can
be taken into consideration. The last part of the paper presents
a workflow for the adequate handling of accuracy, as a function of
a certain noise model. Since the development is focused on the accuracy
of methods instead of expensive hardware elements, it is essential
to know the theoretical capabilities of the system. The focus
of the paper is finding the initial point of edge tracking, the accuracy
of angle calculation and the accuracy of line tracking supposing
a noise model. Finally some experiment results are demonstrated.

**Key words**: 3D scanner, edge tracking, CAD object model, noise model. - Reinoso O., Sebastián J.M., Aracil R., Torres F.:

**Morphological operations with subpixel resolution on digital images**.

MGV vol. 10, no. 1, 2001, pp. 89-102. -
In many defect inspection systems computer vision pattern matching
techniques are applied. These techniques require the use of different
morphological operations. In many situations, pattern images generated
by classical morphological operations do not allow to detect defects
in images. Small defects of pixel size can also be left unnoticed when
using such pattern images. The purpose of this paper is to combine
classical morphological operations with a linear interpolation process
on digital images to generate these pattern images. It is possible
to employ structuring elements of any size to carry out morphological
operations on continuous signals. However, on digital images or discrete
signals, the size of the structuring element should be greater
than pixel size. The approach described in this paper applies classical
morphological operations on reconstructed images at subpixel level
to generate pattern images. In such circumstances, we can compare
this procedure with the effect produced by the use of structuring
elements smaller than pixel size.

**Key words**: morphological operations, pattern matching, sub-pixel accuracy, linear interpolation. - Wang J., Lin Z., Zhu J.:

**A novel recognition method for color objects under different illuminations**.

MGV vol. 10, no. 1, 2001, pp. 103-112. -
Color images depend on the scene illumination, but these
image colors are not stable features for object recognition.
We develop an algorithm that transforms color into its normal
form such that it is invariant to different illuminations.
Since the normalization process is to make color image
compact, colors distorted in different ways from the same
object due to illumination change will all be compacted
to their most compact form and become similar. After compaction
process, the histograms for the same objects under different
illuminations are very similar. The recognition can be
performed by simple matching method. A set of experiments
on complex scenes under various illumination conditions
demonstrate superiority of our proposed method in term
of recognition rate over other reported techniques
for color normalization.

**Key words**: affine transform, color image compaction, covariance, histogram, invariant image matching.

- Kim S.-H., Park R.-H.:

**Fast range image registration using distance sampling**.

MGV vol. 10, no. 2, 2000, pp. 115-131. -
This paper proposes a fast range image registration algorithm,
in which control points are sampled on the basis of distance
from the geometric origin of an object. The sampled point is assumed
to be in the same region if the quantization error in the three-dimensional
space is less than a threshold. For range image registration, finding
matching points only in the same region reduces the computation
time greatly. Experiments with various synthetic and real images show
that the accuracy of registration parameters is improved with
a low computational load.

**Key words**: range image, registration, iterative closest point (ICP) algorithm, quaternion, rotation, translation, sampling and grouping by the distance. - Adhami L., Abdel-Malek K., McGowan D., Sameh A.:

**A partial surface/volume match for high accuracy object localization**.

MGV vol. 10, no. 2, 2000, pp. 133-154. -
A new surface/volume rigid registration method is proposed that allows
fast and robust convergence. In this work, rigid bodies are modeled
using a matrix of voxels and its associated compressed voxel
representation (V-rep). The two models are fused to obtain an indication
of the quality of the match. Optimization techniques using gradient
descent and/or genetics algorithms are implemented to select
the appropriate transformations, where the search space is established.
A sequence of translations and rotations are applied to one of the models
until an appropriate match is reached through this space search.
Unique data structures are introduced to efficiently store the data.
The method and an experimental computer code are demonstrated
in the paper.

**Key words**: surface matching, registration, bone surface identification. - Xu Z., Toncich D., Wu H.:

**Vision-based high accuracy work-piece inspection from profiles**.

MGV vol. 10, no. 2, 2000, pp. 155-174. -
The objective of the research documented herein was to investigate
high-accuracy identification and measurement of a work-piece through
edge detection, and to derive its three-dimensional properties
from a profile of the work-piece image. A number of techniques
for providing the transformation of the image are also described.
The helix, helicoid and range of geometric entities have been selected
to demonstrate the methodology and to highlight various aspects
of the proposed techniques.

**Key words**: vision systems, measurement of geometric properties, edge detection, image analysis, three dimensional shape. - Zeng P.F., Hirata T.:

**Interpolatory edge detection**.

MGV vol. 10, no. 2, 2000, pp. 175-184. -
Conventional edge operators cannot detect fine edges or edges
in low-resolution images in a satisfactory way. In this paper,
an interpolatory edge operator is proposed to deal with fine edge
detection as a supplement. Fast image expansion is implemented
by means of B-spline interpolation. Edges are determined
by seeking pixels with a local maximum of first derivative modulus
(
*modulus maximum*). The first derivatives can be calculated quickly by translation and subtraction along*x*and*y*directions of images interpolated by B-splines. Experimental results imply that the proposed interpolatory edge operator can satisfactorily solve the fine edge detection problem in cases when conventional edge operators fail to produce a clear output. Satisfactory results can be obtained when images are interpolated with the expansion rate of three.

**Key words**: B-spline, image interpolation, edge detection. - Saeed K.:

**Computer graphics analysis: a method for arbitrary image shape description**.

MGV vol. 10, no. 2, 2000, pp. 185-194. -
A concept of a new set of features suitable for describing arbitrary
shapes after their thinning is presented. The features are the minimal
eigenvalues of Toeplitz matrices formed from rational functions,
in which the coordinates of the characteristic pixels of the shape
to be described are used as coefficients, the
*x*coordinates in the numerator polynomial, and the*y*coordinates in the denominator polynomial. Each rational function has a form which makes it a general measure of image description and its features extraction. The method is introduced and applied for the first time.

**Key words**: image recognition, image shape description, computer graphics, feature extraction, Toeplitz-matrices applications. - Serir A.:

**Linear features detection in SAR images**.

MGV vol. 10, no. 2, 2000, pp. 195-205. -
The constant false alarm rate "ratio" edge detector has proved suitable
for synthetic aperture radar (SAR) images. In this paper, we introduce
another ratio detector that uses a half-gaussian weight function
for better localization of linear features, and multiscale functions
for finer detection. A recursive implementation is proposed in order
to reduce the computing time. Tests on 3-view SAR ERS-1 images
as well as on a speckled simulated image are given.

**Key words**: linear features detection, geometric filter, statistics, SAR images. - Konevsky O. L.:

**Smoothing of binary raster images using mathematical morphology**.

MGV vol. 10, no. 2, 2000, pp. 207-220. -
A new method for preparation of binary raster images for curve detection
and subsequent processing is described in this paper. The proposed technique
allows to smooth binary raster images using basic operators of mathematical
morphology in order to eliminate defects and reduce noise while preserving
the topology of the object. The processing parameters are adjusted
automatically to the particular image depending on its features.

**Key words**: mathematical morphology, smoothing, raster image. - Hassanien A.E., Henawy I.E., Own H.:

**Multiresolution image denosing based on wavelet transform**.

MGV vol. 10, no. 2, 2000, pp. 221-230. -
Wavelet-based image denoising is a very attractive tool for analysis
and synthesis of functions. It enables us to divide a complicated function
into several simpler ones and study them individually. In this paper,
we present a new image-denoising algorithm based on multiresolution
local contrast entropy of wavelet coefficients. Depending on the probability
distribution of the noise in the wavelet coefficients, a new adaptive
threshold estimation algorithm is introduced. This threshold enables
the proposed algorithm to adapt to unknown smoothness of denoised images.
The experiments performed confirm that the proposed algorithm is capable
of achieving good results for additive white gaussian noise.

**Key words**: multiresolution, local contrast entropy, denoising, wavelet transform, thresholding, image processing. - Stawowy M.:

**The computer analysis of images in application to the measurement of the vehicles movement parameteres**.

[Dissertation abstract]

MGV vol. 10, no. 2, 2000, p. 231.

Special Issue Editor: Leszek Chmielewski.

- Vincent E., Laganière R.:

**Matching feature points in stereo pairs: a comparative study of some matching strategies**.

MGV vol. 10, no. 3, 2001, pp. 237-259. -
Several algorithms are proposed in the literature to solve
the difficult problem of feature point correspondence between image
pairs. In order to obtain good quality results, they make use
of different approaches and constraints to improve the quality
of the matching set. A matching strategy is considered useful
if it is able to filter out many of the mismatches found in an input
matching set while keeping in most of the good matches present.
In this paper, we present a survey of different matching strategies.
We propose an empirical evaluation of their performance.
The validation process used here determines the number of good
matches and the proportion of good matches in a given match set
for the different parameter values of a matching constraint.

**Key words**: feature point correspondence, stereo matching, interest point detection, corners. - Jan J., Janová D.:

**Complex approach to surface reconstruction of microscopic samples from bimodal image stereo data**.

MGV vol. 10, no. 3, 2001, pp. 261-288. -
The paper describes a set of approaches and methods enabling robust
and relatively precise stereo-analysis-based surface reconstruction
in scanning electron microscopy (SEM). The paper primarily deals
with the disparity analysis problem, namely with selection
of a suitable similarity criterion to be used for finding image
correspondences. The search-and-match method (as opposed
to feature-based analysis) is shown as probably the only practical
solution in the given environment of SEM when no prior constraints
on the surface type are allowed. Extensive comparison of some common
and newly suggested similarity criteria lead to the conclusion
that the designed angle criterion is the only one acceptable so far
with respect to the error rate. Use of the criterion has been shown
equivalent to applying a non-linear two-dimensional matched filter,
which enables efficient frequency domain implementation in the form
of a linear matched filter modification. An important improvement
in reliability of the computed disparities has been achieved by using
both available imaging modalities (back-scattered electrons - BEI
and secondary electrons - SEI), thus providing vector image data.
Expressing the criterion for the vector case in terms of both
individual scalar cases cuts computational requirements by half,
besides allowing for an additional reliability criterion - comparison
of three different, though partly, dependent criteria. Secondly,
the comprehensive approach includes also solutions of problems
which may seem marginal but are important for the practical success
of the analysis. Recent improvements, solving some of such specific
problems of SEM stereo analysis, are discussed as well.
The paper summarises the present state of the method's development
over the past few years partial descriptions of which can be found
scattered in previous publications devoted to individual specific
problems.

**Key words**: surface reconstruction, scanning electron microscopy, bimodal images, image analysis, stereo, disparity analysis, similarity critera. - Cyganek B.:

**Novel feature-based stereo matching method that employs tensor representation of local pixels neighborhoods in images**.

MGV vol. 10, no. 3, 2001, pp. 289-316. -
Stereo matching techniques have evolved substantially throughout
recent years. However, the problem of unambiguous stereo points
matching, especially in presence of object occlusions, as well as
images noise and distortions, remains still open. In this paper,
a novel feature-based stereo matching method, based on tensor
representation of local structures in digital images, has been
described. Application of a structural tensor enables more reliable
matching of locally coherent structures, representing averaged
dominant gradients in local neighborhoods rather than sparse points.
The presented work has been completed with many experiments that
confirmed its usefulness, especially in a case of real stereo images.

**Key words**: stereo vision, stereo matching, depth recovery, structural tensor. - Rziza M., Aboutajdine D.:

**New stereo vision matching algorithm based on constrained dynamic programming and higher order statistics criteria**.

MGV vol. 10, no. 3, 2001, pp. 317-331. -
We present a new efficient stereo algorithm addressing robust
disparity estimation in the presence of noise. We propose here
a new constrained dynamic programming algorithm based
on Higher Order Statistics (HOS) criteria for matching noisy
images. Experiments with both synthetic and noisy real images have
validated our method and have clearly shown the improvement over
the existing ones. The obtained dense disparity map is more reliable
when compared to the similar Second-Order Statistics (SOS) based
constrained dynamic programming and HOS-based correlation methods.

**Key words**: rectification, disparity, matching, correlation, dynamic programming, constrained dynamic programming, higher order statistics, cumulant. - Larabi S.:

**Self-matching of stereoscopic images without camera calibration**.

MGV vol. 10, no. 3, 2001, pp. 333-351. -
In computer vision applications where the calibration object is not
available, it is useful to use an uncalibrated stereoscopic head.
Even in this case, to calculate the three-dimensional structure
of the viewed scene, the stereo matching is considered as the key step
in stereo vision analysis. This paper presents a contribution
to resolve this problem when an uncalibrated stereo rig is involved
in a visual task. We propose an algorithm for self-matching
of stereoscopic images of indoor scenes. Based on projective geometry,
the principal idea of the method is to estimate the epipole position
assuming a set of matched 2D surfaces. A voting approach is used to
select the correct matches which produce the same solution.
In practice, as the stereo images are noisy, we propose a mathematical
analysis of the uncertainty measure. We assume that the vertices are noisy,
and we propagate the effect of this noise in the different stages
of the proposed algorithm. The new version of the algorithm allows
to calculate the region where the epipole point appertains, called
the "epipolar region". The stereo matching algorithm has been tested
on both synthetic and real images, and the number of lines
matched demonstrates the robustness of the geometric method.

**Key words**: stereovision, matching, projective geometry, cross ratio, 2D surface, noise, epipole. - Williams J.G., Blake E.H., Ruther H.:

**Extraction of data from low quality photographs to aid the construction of virtual reality models of archaeological sites**.

MGV vol. 10, no. 3, 2001, pp. 353-368. -
A tool has been constructed to use information extracted from photographs
captured using uncalibrated cameras to fill occlusions which occur
in three-dimensional models of photogrammetrically captured sites.
Capturing the geometry of archaeological sites by photogrammetric means
is relatively expensive and, because of the typical layouts of such sites,
usually results in a degree of occlusion.

The essential philosophy underlying the tool is to segment each occlusion into surfaces which may be approximated using curves, and then use the known geometry in the region of the occlusion to calculate the most probable locations of the junctions of such surface segments. Texture of the surface segments is then applied to the three-dimensional model.

The tool has been applied to occlusions of various configurations that are expected to be typical for archaeological sites, and has been found to deal well with such features and to provide accurate patches from typical data sets. It is also shown that the three-dimensional geometric model is clearly improved by the filling-in of the occlusion.

**Key words**: texture extraction, geometry calculation, occlusion, photogrammetry. - Peer P., Solina F.:

**Capturing mosaic-based panoramic depth images with a single standard camera**.

MGV vol. 10, no. 3, 2001, pp. 369-397. -
In this paper we present a panoramic depth imaging system.
The system is mosaic-based which means that we use a single rotating
camera and assemble the captured images in a mosaic. Due to a setoff
of the camera's optical center from the rotational center of the system
we are able to capture the motion parallax effect, which enables the stereo
reconstruction. The camera is rotating on a circular path with the step
defined by an angle equivalent to a single column of the captured image.
The equation for depth estimation can be easily extracted from the system
geometry. To find the corresponding points on a stereo pair of panoramic
images, the epipolar geometry needs to be determined.
It can be shown that the epipolar geometry is very simple if we are doing
the reconstruction based on a symmetric pair of stereo panoramic images.
We get a symmetric pair of stereo panoramic images when we take
symmetric columns on the left and on the right side from the captured
image center column. Epipolar lines of the symmetrical pair of panoramic
images are image rows. We focuse mainly on the system analysis.
The system performs well in the reconstruction of small indoor spaces.

**Key words**: stereo vision, reconstruction, panoramic image, depth image, mosaicing, motion parallax effect. - Wei S.-K., Huang F., Klette R.:

**Determination of geometric parameters for stereoscopic panorama cameras**.

MGV vol. 10, no. 3, 2001, pp. 399-427. -
This paper proposes an approach to solving the parameter determination
problem for a stereoscopic panorama camera. The image acquisition
parameters have to be calculated under given constraints defined
by application requirements, the image acquisition model,
and specifications of the targeted 3D scenes. Previous studies
on stereoscopic panorama imaging have paid great attention on how
a proposed imaging approach supports a chosen area of application.
The image acquisition parameter determination problem has not yet been
dealt with in these studies. The lack of guidance in selecting
the image acquisition parameters affects the validity of results obtained
for subsequent processes. Our approach towards parameter determination
allows satisfying the common requirements of 3D scene visualization
or reconstruction applications: a proper scene composition in the resultant
images; an adequate sampling at a particular scene distance; and the desired
stereo quality (i.e. depth levels) over diverse of scenes of interest.
The paper details the models, constraints and criteria used for solving
the parameter determination problem. Some practical examples are given
in order to demonstrate the use of the formulas derived.
The error analysis of our determination approach is also carried out
and elaborated in this paper. The study contributes to the design
of stereoscopic panorama cameras as well as to manuals for on-site
image acquisition. The results of our studies are also useful for camera
calibration, or pose estimation in stereoscopic panoramic imaging.

**Key words**: panorama camera, image acquisition, line camera, stereoscopic panorama.

- Cheng B., Li X., Zheng N., Quan W.:

**3D plenoptic representation of a linear scene**.

MGV vol. 10, no. 4, 2001, pp. 431-446. -
This paper presents a novel 3D plenoptic function. We constrain camera
motion to a line, and create a linear mosaic using a manifold mosaic.
The plenoptic function is represented with three parameters:
camera position along the axis, the angle between the ray
and the centric axis, and the rotation angle in the vertical plane.
Novel views are rendered by combining the appropriate captured rays
in an efficient manner at the rendering time. Like panoramas,
our method does not require recovery of geometric and photometric
scene models. Moreover, it provides a much richer user experience
by allowing the user to move freely in a linear region and observe
significant parallax and lighting changes. Compared with either
Lightfield or Lumigraph, it has a much smaller file size because
a 3D only plenoptic function is constructed. Finally, an experiment
with a synthetic environment is given to demonstrate its efficiency
in capturing, construction and rendering of a linear scene.

**Key words**: plenoptic function, virtual environments, image-based rendering, linear mosaic - Kasinski A.J., Hamdy A.M.:

**Segmentation based on homomorphic filtering and improved seeded region growing for mobile robots tracking in image sequences**.

MGV vol. 10, no. 4, 2001, pp. 447-466. -
In this paper, we present a method for extracting and tracking
of mobile robots in a sequence of noisy frames, assuming a complex
background composed of textured floor, illuminated unevenly.
A homomorphic filter is used, as a preprocessor, to enhance
the acquired frames by eliminating the illumination component
and emphasizing the reflectance component of the image function.
To speed up preprocessing of each frame, filtering is only applied
to the pixels belonging to the regions of interest (ROI).
In all the tested cases, homomorphic-filtering led to better
results than those obtained without preprocessing.

The segmentation process has been based on seeded region growing procedure for reconstructing the shape of the mobile robot. We proposed automatic seed points selection in the binarized difference image, and use of an adaptive threshold. This use eliminates or at least considerably reduces false negative detections, and reduces sensitivity of aggregation results to the selected seed points as compared to the classical seeded region growing procedure. Additionally, by imposing a condition of strong connectivity between a seed point and its neighborhood, aggregation of undesired pixels efficiently eliminates false positive detections. Implementation of segmentation and tracking can be run in real time. High tracking accuracy has been obtained through out all the frames in a test sequence.

**Key words**: adaptive threshold, homomorphic filtering, motion segmentation, seeded region growing, visual tracking. - Kim H.S., Yang Y.K.:

**An automatic classification technique for indexing of soccer highlights using neural networks**.

MGV vol. 10, no. 4, 2001, pp. 467-487. -
A method for automatic classification of offensive play patterns
in soccer games has been developed using the neural networks technique.
Back-propagation (BP) neural network techniques have been applied
to obtain data that define the positions of both a player and the ball
on the ground. The offensive play patterns that have been formulated
from the group formations enable automatic indexing of the highlights
of soccer games. Excepts from actual soccer games, including some
from the 1998 French World Cup, yielded 297 video clips which
were categorized into the following five types of patterns:
Left-Running are 60, Right-Running 74, Center-Running 72,
Corner-Kick 39 and Free-Kick 52. Examination of the results shows
the following rates of satisfactory pattern recognition:
Left-Running comes to 91.7%, Right-Running 100%, Center-Running 87.5%,
Corner-Kick 97.4% and Free-Kick 75%.

**Key words**: Soccer game, offensive play patterns, neural network, BP algorithm, group formation. - Otero J., Otero A., Sànchez L.:

**Mode based hierarchical optical flow estimation**.

MGV vol. 10, no. 4, 2001, pp. 489-501. -
In this paper, a robust estimation of the optical flow field that preserves
the boundaries of the movement is shown. Arising from the techniques based
on the Optical Flow Constraint (OFC), an estimation that takes several
measures around a given pixel, discarding the erroneous ones, has been
developed. This is done through performing a bidimensional clustering
of the velocities obtained from the intersection of pairs of OFCs.
In this way, the clustering is conducted in the velocity space
and not in the (slope, intercept) parameter space of the OFCs.
Finally, a hierarchical implementation that has a lesser error
when large displacements are present in the image is shown.

**Key words**: computer vision, motion analysis, optical flow, robust estimation. - Jun Wei H., Lei G.:

**An algorithm for automatic detection of runways in aerial images**.

MGV vol. 10, no. 4, 2001, pp. 503-518. -
This paper presents an automatic algorithm to detect runways
of the military airport in aerial images. Firstly, we design
a model of runway based on its features. Then, we find
the runway in a hypothesis and test paradigm. Hypotheses
are formed by looking for instances of long rectangular
shapes. Runway intensity and intensity contrast between
the runway and background are used to verify our hypotheses.
Finally, the hypothesis that is supported by verification
is believed to be the runway. Many experiments demonstrate
that this algorithm can find runways of a military airport
effectively.

**Key words**: runway, edge, edge tracking, straight line, anti-parallel. - Grabska E., Slusarczyk G., Niewiadomska A.:

**Hierarchical graph grammars in graphic prints design and generation**.

MGV vol. 10, no. 4, 2001, pp. 519-536. -
The paper deals with graphic prints in Escher's style
and hierarchical graphs. Escher prints are based
on regular plane divisions, while hierarchical graphs
represent plane division structures. The paper is illustrated
by graphic prints generated by the system
*ESCHER_GRAPHICS*.

**Key words**: graph, graphical model, graph grammar, graphic print. - Iwanowski M.:

**The metamorphosis of 3D binary objects using morphological interpolation**.

MGV vol. 10, no. 4, 2001, pp. 537-550. -
This paper presents a new method of metamorphosis of three-dimensional
objects by means of an interpolation method based on mathematical
morphology supported by an affine transform. The metamorphosis
is a process of smooth transformation of one object into another,
the result of which is stored in a sequence of 3D images.
The proposed method is automatic - the only data necessary
consists of: two input 3D binary objects (the initial one
and the final one) and an additional parameter - either
the number of frames in the final sequence or the highest
value of the dissimilarity measure. The method consists
in performing successive computations of morphological median
objects. The proposed interpolation scheme is a new application
of morphological interpolation to 3D image metamorphosis.
Also, it combines 3D morphology with affine object matching.
The proposed interpolation sequence algorithm is a new one
and allows us to produce the interpolation sequence
by considering dissimilarity measures between the consecutive
frames. The method can be applied to animation in computer
graphics and visualization of three-dimensional data.

**Key words**: 3D objects interpolation, volume morphing, mathematical morphology, morphological interpolation. - Revievers' index
- Authors' index
- Contents of volume 10, 2001

9 (2000) | main | 11 (2002) |