|
This course will introduce students to the essential
techniques for using computers to extract useful information from
photographs, 3D surface scanners, and volumetric medical images. If you have any questions, email me:
first initial, then last name, at ucdavis.edu.
Who can take it?
Undergrads and grad students are
welcome to take the class. I am
going to try like heck to make it not too basic for grad students, but not
too hard for undergrads either.
Prerequisites are knowledge of data structures and programming
techniques, plus a background in linear algebra. If you have taken courses in this, great; if you know
programming and linear algebra through some other avenue, let me know and
weÕll talk about it.
Topics
|
|
2D images: We
will talk about algorithms for automatically extracting information out of
digital photographs and movies.
We will discuss the basics of how to detect basic features such as
edges, corners, and visual patterns (such as stripes, leopard spots, etc),
and how to use those basic features to detect objects like this
ladder. We will also talk
about how to use information from multiple images to build 3D models of
objects and environments.
|
|
|
3D surface imagery:
Here we will focus on data that comes from 3D surface sensors such as RADAR
and laser range finders: they typically give you a set of 3D points that
lie on the surfaces they are pointed at. We will talk about how to remove noise and other
artifacts from surface data, how to detect objects in 3D data, and how to
combine multiple views of the same object into a single, unified 3D model
of it (like this plumbing part).
|
|
|
Volumetric medical images: Medical imaging modalities such as CT, MR, PET
and SPECT give you a filled-in, solid 3D ÒcubeÓ of data that represents the
physical properties of molecules inside part of the human body (the brain,
for example). We will talk a
little about how these images are formed and how to get rid of artifacts
from them. We will then
discuss how to automatically detect that an object is present in one of
these images—for instance a brain tumor. We will go into techniques for mathematically describing
the shapes of biological objects (like the brain) and how the shapes change
over time and over the course of disease. For example, the image to the left depicts changes to a
key brain structure as a result of AlzheimerÕs disease.
|
Key components of the course
|
|
Lectures:
Students will learn algorithms for automatically detecting objects in
images of various types, building 3D models of objects and scenes from
example views, removing image artifacts, and mathematically representing
the shapes of things. I will
try to give students a general sense of what the Òbig problemsÓ in image
processing are, and the key characteristics of available solutions.
|
|
|
Programming assignments: Students will program image processing algorithms in small teams
using the Insight Toolkit (ITK), a cross-platform, open-source package of
C++ classes and supporting software.
ITK is currently the industry standard for public-domain image
processing functionality, especially for medical images. Most of the course grade (60%) will
be based on the results of 3 programming assignments, in which students use
ITK to add significant functionality to bare-bones code that solves a major
problem for 2D images, 3D surface imagery, and medical images.
|
|
|
Exams: There
will be a midterm and final, corresponding to 15% and 25% of the grade,
respectively.
|
|
