Seminars at CBA, Spring 2005
( Please send your e-mail address to me, if you wish to join our seminar-reminder e-mail list )
Kvalitetskontroll av flis för massatillverkning med datoriserad bildanalys
Skoghalls pappersbruk tar delar av som råvara från sågverk i form av flis. Eftersom
massaprocessen är känslig för störningar såsom bark och annat skräp måste fliskvaliten
kontrolleras då flisen kommer till pappersbruket. Idag görs detta med manuella stikprover där
man bl a kontrollerar andelen bark mätt i volymprocent. Det här exjobbet skulle undersöka om
detta är möjligt att göra med bildanalys.
Två olika bildserier har använts. Bildserie ett för att utveckla metoden och bildserie två
för utvärdering. Resultaten indikerar att det är möjligt att skilja ut barken från övrig flis
med bildanalys. Bilderna från bildserie två ger ett sämre resultat an bildserie ett, till
stor del beroende på att nya bitar av bark är representerade i bildserie två, vilka metoden
inte känner igen. Genom att utveckla metoden mer med fler bilder finns möjlighet att
A medial grey-level based representataion for molecules in volume images
I will present an algorithm to extract a medial representation of molecules in volume images. The representation takes into account the internal grey-level distribution of the molecule and no "true" segmentation of the molecule is necessary prior to extracting the representation.
Digital lines with irrational slope - theory
Hania Uscka - Wehlou
I will present the results of my research about digital lines
with irrational slope. It will be the formal description of such lines. I will
present the theory needed for the formulation of necessary and sufficient
conditions to be a digital line with irrational slope. There is no advanced
mathematics involved, but quite a lot of definitions are needed for the
final formulation of the conditions.
Object segmentation in color images
(The seminar will be held in swedish)
Digital cameras are gaining in popularity, and not only experts in image
analysis, but also the average users, show a growing interest in image
processing. Many different kinds of software for image processing are
available, but most of them require expertise knowledge, and are too advanced
for the average user.
Object segmentation is the process of finding, outlining, and extracting objects
in an arbitrary digital image. Image processing software often provide a set of
different tools for segmentation. Many of the tools are, however, too
complicated for the average user, or leave too little freedom in expressing the
The aim of this master thesis work was to develop an easy to use tool for object
segmentation in color images for the average user. The work resulted in a new
color image segmentation method with little user interaction and no tuning
parameters. The method is based on the Watershed segmentation algorithm,
combined with seeding information given by the user, and color space
projections for optimized object edge detection. The presented method can
successfully segment objects in most types of color images.
The Euclidean distance transform applied to the fcc and bcc grids
The discrete Euclidean distance transform is applied to grids with non-cubic voxels, the face-centered cubic (fcc) and body-centered cubic (bcc) grids. These grids are three-dimensional generalizations of the hexagonal grid. Raster scanning and contour processing techniques are applied using different neighbourhoods. When computing the Euclidean distance transform, some voxel configurations produce errors. The maximum errors for the two different grids and neighbourhood sizes are analyzed and compared with the cubic grid.
Haptic volume rendering based on gradient vector flow
In this talk I will present a haptic volume rendering technique based on gradient vector flow. The aim is to facilitate the manual part in semi-automatic segmentation, e.g., placement of seedpoints and/or positioning of surface models inside objects to be segmented.
The method allows a user to stay centered inside the object while still feeling the object boundaries. Initial tests of the method shows encouraging results for differently shaped objects.
Mean shift filtering
In an arbitrary image there is a lot of local, and global, variation between the intensities stored in the different image elements. Sometimes this variation is a wanted feature, but many times the image is subsequently segmented, with the aim of reducing the variation into regions. Mean shift filtering is a mean based algorithm that reduces variation in the image. In a closer look the filtering is an adaptive gradient ascent method, with discontinuity preserving properties. The filtered image lends itself to many segmentation algorithms with interesting result.
On the language of discrete planes and surfaces
We introduce a generalization of the notion of a standard
discrete plane, namely the (1,1,1)-discrete surfaces. We first study a
combinatorial representation of discrete surfaces as two-dimensional
sequences over a three-letter alphabet and show how to use this
combinatorial point of view for the recognition problem for these discrete
surfaces. We then apply this combinatorial representation to the standard
discrete planes and give a first attempt at generalizing the study of the
dual space of parameters for the latter.
Suspended Matters Impact on Color Reconstruction in Underwater Images
The natural properties of water column usually affect
underwater imagery by suppressing high-energy light. In application
such as color correction of underwater images estimation of water
column parameters is crucial. Diffuse attenuation coefficients are
estimated and used for further processing of underwater taken data.
The coefficients will give information on how fast light of
different wavelengths decreases with increasing depth. Based on the
exact depth measurements and data from a spectrometer the
calculation of downwelling irradiance will be done. Chlorophyll
concentration and a yellow substance factor contribute to great
variety of values of attenuation coefficients at different depth. By
taking advantage of variations in depth, a method is presented to
estimate the influence of suspended matters on color correction.
Attenuation coefficients that depends on concentration of suspended
matters in water gives an indication on how well any spectral band
is suited for color correction algorithm.
Converting a digital camera into a spectrometer
Hamed Hamid Muhammed
I'll present a novel approach for modifying an ordinary digital camera to be able to generate multi- or hyperspectral images. The basic idea is to use miniature (spatially and spectrally) overlapping color-mosaic filters with a spatial resolution which is as close as possible to the actual resolution of the sensor plate. What is new here is that common (cheap) printing techniques can be used to produce these filter mosaics. The resulting image which shows locally filtered scene areas, can be transformed into a hyperspectral image by considering each group of neighboring pixels and transforming it into a single image element in the final hyperspectral image. The new technique has been developed by me and Fredrik Bergholm.
Gas jet impinging on liquid surface: Cavity shape modelling and video-based estimation
A water model is studied to simulate physical phenomena in the LD
steel converter. The depression in the liquid, due to the impinging gas jet, is measured by means of a video camera. Image processing tools together with a nonlinear mathematical model based on the physics of the liquid-gas system, are used to describe the cavity profile. The properties of the model are investigated. A quantification of the uncertainty of the cavity depth estimates is given and the frequency content of the oscillations of the indentation profile is characterized by using the Fast Fourier Transform.
Analysing the void space of porous material (for example paper)
The pore structure of paper is of interest for the paper quality, as it
affects how the paper interacts with light, fluid, and air, as well as its
mechanical properties. Recently, we (Ida, Maria, Gunilla & I) have developed
a method for partitioning the void space of paper into individual pores. I
will briefly explain this method. Moreover I will talk about how "pore
people" usually describes the void space, what type of information that is
of interest to extract and give some idea how this can be done using image
Introduction to CVS
When writing programs, both small and large, it is useful (not to say
vital) to keep a history of the changes made to the files. CVS is a
version control system which keeps track of all the changes made to a
program during the entire development time. This provides the developer
a safety net.
I will give a brief introduction to CVS, and how to use it.
Mathematical Morphology for the Classification of Remote Sensing Images from Urban Images
In this seminar, the classification of panchromatic high-resolution remote
sensing images from urban areas is considered. Panchromatic
high-resolution urban images are difficult to classify using conventional
pixel based classification approaches. Actually, since such data only
consist of one single channel, spatial information is more important than
spectral information. It is therefore necessary to use geometric
information to improve the pixel based classification results.
Consequently we propose to use mathematical morphology to extract some
characteristic shape features on which the classification can be based.
Regardin the classification step, an artificial network has been tested,
as well as a fuzzy possibilistic model and the fusion of these two
approaches. This work was conducted with Jon Atli Benediktsson at the
University of Iceland and Mathieu Fauvel (PhD student).
Lite om University of Queensland och min forskning där
This seminar will be a rather informal one
about my sabbatical in Brisbane and
my research there. The scientific part
will be rather similar to the paper presented
at WDIC in Brisbane in February and
at SSBA in Malmö in March dealing with
"Dynamic breast MRI visualised through colour mapping."
Spatial and spectral analysis of fluorescing blobs
Spatial and spectral analysis of fluorescing blobs Since the first of
January, I have been working half time at the department of Genetics and
Pathology, the group of molecular medicine/molecular tools. The group is
developing methods and reagents that enable localized detection of genes and
transcripts at single-nucleotide resolution and at single-molecule
Sensitive methods for analysis of subtle DNA and RNA sequence variants in
vitro and in situ are of great importance in applications ranging from the
study of tumor progression and genetic disease to fast evaluation of
drinking water and microbes used in biological warfare.
The output from the detection methods is mostly digital images of
fluorescing "blobs" produced by confocal microscopy or multi spectral
confocal microscopy. I will present some of the methods for pre-processing,
segmentation, automated thresholding, and calibration that I've been working
with the last couple of months.
Imaging paper with X-ray microtomography
At the European Synchrotron Radiation Facility, ESRF, in Grenoble
it is possible to make X-ray microtomography images of paper. Usually the
internal structure of paper is not possible to image with good enough
resolution in a normal lab, but the resolution at ESRF is far below the
required 1 um. At the seminar I will descibe the methods used to image the
samples and show some if the resulting volumes.
Object Detection and Tracking in Video
Prof. Ranga Kasturi
Dept. of Computer Science and Engineering
University of South Florida
Tampa FL, USA
In this talk I will present an overview of several research
projects that we have completed during the past few years. These include
text detection in broadcast video, collision avoidance for aircraft
navigation, human motion trajectory tracking, multimodal biometrics, and
performance evaluation of video object detection and tracking algorithms.
Bio: Rangachar Kasturi received his degrees in electrical engineering from
Bangalore University (B.E., 1968) and Texas Tech University (M.S. 1980 and
Ph.D. 1982). He was a professor at the Pennsylvania State University from
1982-2003. He is now a professor and the chairman of computer science and
engineering at the University of South Florida. He was the president of
IAPR during 2002-04 and was the editor-in-chief of IEEE Transactions on
Pattern Analysis and Machine Intelligence during 1995-98.
A New and more Appropriate Application of Principal Component Analysis for Improvement of Image Quality and Clinical Diagnosis of Human Brain in PET Studies
Dynamic PET data acquisitions are usually used in studies on human brain, where whole brain is sequentially imaged at different time points or frames. In other words in these types of acquisition, data sets are generated in the form of similar images obtained from the same scene from different frames. This type of data acquisition is used for providing images containing physiological, biochemical and functional information of human brain. This information can be derived by analyzing the distribution and kinetics behavior of administrated tracers in different regions of the brain. Each one of these images contains part of kinetic information of administrated tracer within the time sequences.
Two disturbing factors which make the analysis of PET data difficult independent of the used tracer, yet potential to estimate precision in a measurement, are high level of noise magnitude and correlation of noise between the pixels within image. One of the standard methods used for reduction of the noise and quantitative estimation in dynamic PET data is to sum images. The summation usually applies on part of the sequence where the specific signal is proportionally larger. One of the drawback using sum images for analysis of PET data is that summation of images tends to dampen the differences between regions with different kinetics. Anther method used for analysis of dynamic PET data is kinetic modeling such as reference Patlak, which are used for generation of parametric images, aiming to extract areas with specific kinetic properties that can enhance the discrimination between normal regions contra pathology. A severe problem using kinetic modeling is that the generated parameter images suffer from poor quality predominantly while the images are rather noisy.
Principal component analysis (PCA) is one of the most commonly used multivariate images analysis tool for analyzing dynamic PET data. PCA has been used in order to find variance-covariance structures of the input data aiming for reduction of dimensionality of the data set. It has been shown that since PCA is a data-driven technique it could emphasize the noise over of regions with different kinetic if input data is not properly handled/pre-normalized for variable noise that exists in the input images. To overcome this problem we introduce a new and proper application of PCA on different dynamic PET studies on the brain. Here we have used different tracers, to explore the performance of the new method in order to improve detection, visualization of significant changes of tracers kinetic and enhancement of the discrimination between pathological and healthy regions in the brain.
Creating Quality Imagery and Video from Inexpensive UAV
- through determining Camera Motion and 3D Structure
Harald Klomp & Jakob Sandström
Due to the small size and limited payload capability of mini-UAV systems
of sensors is often not a feasible option. Turbulence is often significant
at the typical operating height
of most mini-UAV systems and cost requirements usually dictate the use of
non-metric cameras. By
consequence the raw imagery acquired is often characterized by unstable
orientation, motion blur,
high noise levels and significant optical distortions. Hence, extensive
processing has to be applied in
order to meet the user requirement of crisp stabilized video and high
resolution image mosaics with
map compatible geometry.
The focus of this thesis is to explore methods and techniques used in
Computer Vision and Pho-
togrammetry in order to retrieve geometric information and camera motion for
We propose a simple linear stabilization of the video, compensate for the
optical distortion introduced
by the cameras used and propose SIFT feature detection, RANSAC feature
matching and global bun-
dle adjustment as the solution to the problem of extracting geometric
information from an unordered
sequence of aerial imagery.
Please send your e-mail address to me, if you wish to join our seminar-reminder e-mail list
Hamed Hamid Muhammed
Last modified: Wed May 18 11:34:43 MEST 2005
. . . . . . . .