A. Course objectives and preliminaries
This is the website for my Fall 2009 seminar on computer vision.
The purpose of the seminar is to explore (take a guess) current
problems being addressed by the computer vision community: those
issues, approaches and applications
evoking notable current research interest and promise. At the end of
the seminar, participants should be aware of major trends and
directions in
current computer vision research, have selected one to explore in some
depth, and given a tutorial presentation on it.
The first two seminar meetings will be devoted to surveying the various
topics to be discussed throughout the rest of the semester. Each
registered student will
select one of these topics (or suggest a different
topic which I approve), and deliver a presentation on that
computer
vision research topic based on a literature search of work in the area.
The goal of each presentation is to present a tutorial description of
the problems and methods associated with the selected topic, and the
presenter's thoughts concerning the most important unsolved problems
and future directions of research in the area.
This
website will serve mainly as a
repository of files, links and other information relevant to the
course. As student presentations are completed, they will be added to
the website. The
presentation, and attendance at most/all of the other student
presentations, are the only course requirements for students
registering for 1 credit. For those registering for 2-3 credits, at
the end of the semester, each such student will submit a research
paper, similar in scope, length and scholarship to a paper tendered to
a good conference in the field. The paper should be built on the
presentation, adding some original research ideas and results to the
tutorial material of the presentation. As is the custom in our
department, seminars will be graded S/U unless otherwise letter grades
are requested on an individual basis.
Enrollment is limited to grad students with at least one previous
course in image processing or computer vision. The group will meet
weekly for one 2 hour session at a time of the week that fits
into everyone's
schedule. On the first day of class for the Fall semester, Monday Aug
31, I will email everyone
who has registered for the course and ask their class and teaching
schedules. After perhaps one or two iterations as students add and
drop, I will then send a final email around indicating what our
regular meeting time and place for the semester will be. If you are
interested in attending but do not register for the course, please
check my office door (Bell 136), I will post the time and place for our
meetings as soon as this information is available.
I
will have office hours in my office Bell 136 this semester
2:00-3:00pm Mon and Tues, 3:00-4:00pm Wed. I can
be reached by email at peter@buffalo.edu or phone at 716-645-3180 x
137.
Discussion Q&A:
Can
every
message be represented by a bitstream?
This is the current syllabus,
with a helpful introductory reference for each topic linked. Registered
students are
asked to review
these references before
the class in which they will be surveyed. These
references are just basic tutorial introductions, which give the
flavor of the topic but should not be expected to supply detail and
technical information adequate for a thorough understanding. What they
will do is facilitate good discussion by grounding each seminar
participant in the basic tutorial material before the session in which
they will be discussed.
1. Image processing from still images
1.1 Segmentation: fast marching http://math.berkeley.edu/%7Esethian/Explanations/fast_marching_explain.html
Select Fast Marching Methods from frame.
1.2 Shape recovery: level sets http://math.berkeley.edu/%7Esethian/Explanations/level_set_explain.html
Select Level Set Methods from frame.
1.3 Segmentation: graph cuts http://www.cis.upenn.edu/~jshi/GraphTutorial/Tutorial-ImageSegmentationGraph-cut1-Shi.pdf
Select Part 1.
1.4 Registration: image transformations http://www.cs.wright.edu/~agoshtas/CVPR04_Registration_Tutorial.html
Select Transformation Functions
1.5 Registration of images with maps Haala_2003.pdf
1.6 JPEG2000 Impoco_2004.pdf
2. Computer vision from still images
2.1 OR: SVM methods http://www.support-vector.net/icml-tutorial.pdf
2.2 OR: Learning LeCun_2004.pdf
2.3 Free-viewpoint rendering http://iphome.hhi.de/smolic/docs/Smolic_ICIP04_FVV.pdf
2.4 Multi-view 3D reconstruction Vogiatzis_2005.pdf
2.5 Multi-sensor image fusion http://www.ece.lehigh.edu/SPCRL/IF/image_fusion.htm
Select
Section
1:
Introduction and
Section 2: Review of Image Fusion Research
3. Video processing
3.1 Particle filters for tracking Li_2004.pdf
3.2 MPEG-4/H.264 video coding http://www.m4if.org/resources/IEEESpectrum/mpeg-4.htm
3.3 Model-based tracking http://csdl2.computer.org/comp/proceedings/ismar/2003/2006/00/20060313.pdf
3.4 Tracking using Bayes' nets Murphy_2003.pdf
3.5 Free-viewpoint video http://www.merl.com/reports/docs/TR2003-137.pdf
3.6 Motion detection http://lear.inrialpes.fr/people/triggs/events/iccv03/cdrom/iccv03/0734_viola.pdf
3.7 Motion segmentation Weiss_2003.pdf
4. Computer vision from video
4.1 Detecting surprises, abnormalities Koller_2002.pdf
4.2 Action Recognition Bregonzio_2009.pdf
4.3 Detecting, tracking people Cutler_2003.pdf
4.4 Wayfinding for the vis impaired Nagarajan_2003.pdf
4.5 SLAM: Simultaneous Localization and Mapping Hiebert-Treuer_2007.pdf
5. Mathematical foundations
5.1 Representation: wavelets http://www.amara.com/IEEEwave/IEEEwavelet.html
5.2 Superresolution http://www.ece.rice.edu/~rwagner/docs/wagnerSPIE04.pdf
5.3 Learning Weng_2000.pdf
5.4 Modelling uncertainty: prob, belief, fuzzy, rough sets, intervals, etc. Smets_1999.pdf
5.5 Evaluation methods Martens_2002.pdf
5.6 Mosaicking Hsieh_2003.pdf
5.7 A contrario detection methods VISTA_2004.pdf
5.8 Belief and image uncertainty Sun_2003.pdf
6. Architectures and hardware
6.1 FPGA architectures http://www.us.design-reuse.com/articles/article10943.html
6.2 Vision on a chip http://www.nasatech.com/Briefs/Feb00/NPO20449.html
6.3 Smart cameras Leeser_2004.pdf
7. Miscellaneous
7.1 Watermarking Perez_200X.pdf
7.2 Image steganography Johnson_1997.pdf
7.3 Inpainting methods http://www.math.ucla.edu/~imagers/htmls/inp.html
8. Topics which will not
be discussed
(duplicates other courses)
1. Face recognition
2. Image indexing or retrieval
3. Biometrics (eg. fingerprints)
4. CV for bioinformatics
5. Active or animate vision
Class Schedule
Student presenters and their topics will be filled in as they become
scheduled.
|
Week |
Presenters |
Topic |
Prep Readings |
|
1 08/31 - 09/04 |
|
No
meeting - set up schedule |
|
|
2 09/07 - 09/11 |
Peter Scott |
Overview of Topics I |
|
|
3 09/14 - 09/18 |
Peter Scott |
Overview of Topics II |
2.1-7.3 |
|
4 09/21 - 09/25 |
Kedar
Sarmalkar |
||
|
5 09/28 - 10/02 |
No seminar meeting |
Prof
at
a
conference out of town |
|
|
6 10/05 - 10/09 |
Balakrishna
Thiagarajan Niranjan Kamat |
Graph
Cut Segmentation Tracking via GMMs |
Sharon, Thiagarajan Grest, Stauffer |
|
7 10/12 - 10/16 |
Santhosh
Kandalu Qiang Gao |
Motion
detection 3-D Terrain Following |
Shahinfard, Lee Bors-1, Bors-2 |
|
8 10/19 - 10/23 |
Ning
Zhang Lynn Lobo |
Face
Detection Learning |
Tu, Viola PCA, LDA, Regr. Trees |
|
9 10/26 - 10/30 |
Mukul
Apte Bhushanshirish Chitte |
Super-resolution Watermarking |
Park, wikipedia Provos |
|
10 11/02 - 11/06 |
No seminar meeting |
Presenter health problems |
|
|
11 11/09 - 11/13 |
Ranjan
Seetharama Porchelvi Vijayakumar |
Wayfinding
for the visually impaired Bayesian learning |
Nagarajan, Heyes wikipedia, Gopnik |
|
12 11/16 - 11/20 |
Vikas
Choudary |
CAPTCHA |
Wikipedia, Chew |
|
13 11/23 - 11/27 |
No class |
Fall recess |
|
|
14 11/30 -
12/04 |
Group discussion | Seminar summary and wrap-up* | Nagel, Piccardi |
* In preparation for this meeting, please read the Nagel paper What It Is Like To Be A Bat and the short Piccardi review article, Recent Advances in Computer Vision, and review your own presentation. At this final meeting we will do three things:
1. Go around the room and get everyone's
summary of what they posted concerning the completeness of bitstreams
over the space of messsages, and whether their view has changed since
then, in light of further reflection, other posted views, or the Nagel
paper.
2. Ask each seminar member to identify one research idea of their own on
the topic of their presentation. Hopefully, these are not ideas
contained in one of the references each presenter used, rather each
idea contains some degree of novelty. If you are actively working
on one now, describe what you are doing.
3.
I will summarize the current state of work in computer vision
research as I see it, and as we studied it in this seminar, comparing
my view to that of Piccardi. I will try
identify where I believe the most consequential research is occuring
and can be started.
Unlike our previous meetings, attendance at this last
meeting is mandatory. This is the meeting where we "pick the fruit"
grown during the course of the seminar.