 |
|
Computer Vision
Awards
Kay H. Brodersen received a Trainee Abstract Award at HBM 2012 for his work on 'Model-Based Clustering Using Generative Embedding.' Kay will be giving a talk on his results in Beijing on 12 June.
François Cellier received the McLeod Founder's Award of the Society for Modeling and Simulation International.
Spotlight
Computational models of brain connectivity, coupled with machine learning algorithms, make it possible to infer neuronal disease mechanisms from
non-invasive functional magnetic resonance imaging (fMRI) data in humans.
This illustration shows how dynamic systems models can be used for reducing
complex (high-dimensional) brain activity data to a simple (low-dimensional)
and mechanistically interpretable representation (Brodersen et al., PLoS
Comput. Biol. 2011). Please also see the summary on ETH Life.
Introduction
|
Recent advances in image and video capture have enabled very cheap acquisition
of image data. This calls for automatic tools for analysis, categorization and
interpretation. Such challenges in the field of computational vision present an
ideal test bed for pattern recognition and machine learning algorithms. We apply
new sampling techniques to image segmentation, investigating the different
approaches for efficient large scale inference. Graphical models are used to
learn object categorization from weakly labeled data, providing a framework for
integrating multiple sources of information. We use psychophysics to investigate
object detection in humans and we apply these techniques to real world problems,
such as the analysis of medical and neuroanatomical data.
|
|
Projects
Computational Pathology
|
The great success of histology in general and tissue microarrays (TMA)
in particular is closely related to experimental techniques which yield
localized protein expression values in tissue. High resolution scanning
technologies in recent years renders an automated analysis of
histological slices feasible both for medical research and for
practical clinical work. This project aims at the development of a data
processing pipeline which models the complete pathological workflow
from the microscopic inspection of tissue samples to the diagnostic
analysis of survival curves.
|
|
|
|
|
|
|
|
|
|
|
Collaborators: University Hospital Zürich
|
|
|
3D reconstruction of neural images
|
Goal of the project is to develop image processing and machine
learning methods and algorithms to detect and classify and reconstruct
neuroanatomical structures like neurons, dendrites and synapses from
transmission electron microscopy images. We work in close cooperation
with the Institute of Neuroinformatics UniZH | ETH Zurich (INI) and
electron microscopy ETH Zurich (EMEZ). In order to get a better
understanding of learning and remembering in the brain, neuroanatomists
build 3d reconstructions of brain tissue from electron microscopy
images. Today most of this work involves time consuming manual image
processing, which renders the process very tedious and time consuming.
Automating the image processing pipeline will not only make this
process faster, but enable the anatomist to look at much larger
datasets, hopefully leading to new insights of the functional structure
of the brain.
|
|
|
|
|
|
|
|
|
|
|
|
TEM image of neuronal tissue. Image courtesy of Kevan Martin and Nuno Miguel Macarico A. da Costa (UniZH | ETH Zurich)
|
|
|
|
|
|
|
|
|
|
|
|
Image processing workflow for 3d reconstructions of serial section TEM images.
|
|
|
|
|
|
|
Collaborators:
|
|
|
Discriminative methods for structured data
|
|
|
|
|
|
|
|
Conditional Random Fields allow for a principled, probabilistic
discriminative reasoning about structured data. We are investigating
applications in computer vision like learning how to tag images (top) or
training segmentations of photos into semantic classes (bottom).
|
|
|
|
|
|
|
Utilizing multiple data sources by information fusion
|
This project is dedicated to integrating approximate geometry inference
and object segmentation and recognition. The main challenges lie in
constructing a feature transform that is adequate to both tasks and
devising a learning algorithm that can exploit the relation of tasks
that are different, but defined on same domain.
|
|
|
|
|
|
|
|
Wichtiger Hinweis:
Diese Website wird in älteren Versionen von Netscape ohne
graphische Elemente dargestellt. Die Funktionalität der
Website ist aber trotzdem gewährleistet. Wenn Sie diese
Website regelmässig benutzen, empfehlen wir Ihnen, auf
Ihrem Computer einen aktuellen Browser zu installieren. Weitere
Informationen finden Sie auf
folgender
Seite.
Important Note:
The content in this site is accessible to any browser or
Internet device, however, some graphics will display correctly
only in the newer versions of Netscape. To get the most out of
our site we suggest you upgrade to a newer browser.
More
information
