ENGN4522/6520: Statistical Pattern Recognition and Its Applications in Computer Vision.

The 2^nd semester (21/Jun/2008 --- 31/Oct/2008).

Assignment Four (please download the data set needed in this assignment)

Assignment Two (please download the data set needed in this assignment)

Assignment Three and paper [1] and [2] (The data set is same as that in Assignment Two)

! Important

Please download the Instructions for completing and submitting the assignment reports

Tutorial 2 will be given at the Graduate Teaching Room in Ian Ross building

from 10AM-12PM on 25/Sep/2008

Tutor: Mr. Cong Phuoc Huynh ( huynh@rsise.anu.edu.au )

New courses for 2007 and 2008:

ENGN8531 Advanced Computer Vision and Pattern Recognition

Course Description

Dear students,

We are going to give a course on patter recognition and computer vision. The

schedule of this course is attached below. We want to learn your opinions about

the content, for example, anything else that you think should be included. By

including your suggestions, we hope to make this course more attractive and successful.

Any comments are warmly welcome and appreciated.

Lei Wang (Lei.Wang@rsise.anu.edu.au)

Hongdong Li (hongdong.li@rsise.anu.edu.au)

The course ENGN8531 will be delivered from 21/Jul/2008 to 26/Sep/2008 and

then from 13/Oct/2008 to 31/Oct/2008. There will be one lecture each week and

it lasts 2 hours. This course aims to provide students with modern pattern recognition

techniques and their applications to the field of computer vision. It consists of

two parts. The first part is about the fundamental pattern recognition theories and

approaches. The second part includes a series of talks, each of which focuses on

a particular pattern recognition approach and its applications to computer vision.

The main references for this course include the book .Patter Classification. (2nd

edition) authored by Duda, Hart, and Stork, as well as a number of papers selected

from the top journals and conferences in the field of computer vision. The score

will be given based on the attendance, assignments, and the final project.

The course schedule is listed below. Those marked by .*. are a series of talks.

1. Introduction of pattern recognition (0.5 hour)

2. Bayesian decision theory (1.5 hours)

3. Maximum-likelihood and Bayesian parameter estimation (2 hours)

4. Nonparametric techniques (ParzenWindows and K-Nearest-Neighbor Estimation)

(2 hours)

5. Unsupervised learning (K-means clustering, Hierarchical clustering) (2 hours)

6. EM algorithm, Gaussian Mixture Model, and Hidden Markov Model (2

hours)

7. Linear discriminant analysis (2 hours)

8. * Kernel methods and the applications to Computer Vision (2 hours)

9. Stochastic methods and nonmetric methods (2 hours)

10. Algorithm-independent machine learning (No Free Lunch Theorem, Ugly

Duckling Theorem, Occam's Razor,. . . ) (2 hours)

11. * Boosting techniques and the applications to Computer Vision (2 hours)

12. * Component analysis methods and the applications to Computer Vision (PCA,

LDA, ICA, MDS, ISOMAP, LLE,. . . ) (2 hours)

Literature Reading in Computer Vision (ENGN8532)

Course description

This is a literature-reading-and-presentation course. It is designed to broaden and extend the knowledge base of students in the most recent developments in computer vision and pattern recognition. Besides vision, the topics also cover image processing, graphics, machine learning and pattern recognition. It serves as well the purpose of enhancing the students' scientific communication and presentation skills, and nurturing good research habits.

Detailed information

This course will look at advanced topics in advanced computer vision. Each week, we will read and discuss 1 ~3 papers. The papers are mainly selected from recent top international computer vision conferences such as ICCV, ECCV and CVPR, but may also from major conferences from other related fields, such as SIGGRAPH, NIPS, MICCAI, ICPR, ICIP, ICASSP etc.

Pre-requirement

Programming experience and familiarity with linear algebra and calculus is assumed. Some background in computer graphics, computer vision, or image processing is helpful.

Meeting Time and Place

Thursday 4:00PM - 5:00PM, RSISE- A207 Video Room.

Important date:

Enrolment date: within first two weeks of S-2.

Starting date: 20 ^th of September, i.e, the 4-th teaching period.

End date (final project report and presentation due date): 22^st and 29^th of November.

Instructor

Dr. Hongdong LI, hongdong.li@anu.edu.au

RSISE, Room 333, BLD-115, ANU. Please send me email if relate to the course.

Recommended Texts

Computer Vision: The Modern Approach, Forsyth and Ponce

Assignments (and grading criteria)

I. Reading and class-presentation component

( 40 %):

You will be asked to read carefully one or two papers for each class, chosen from the following list. Each week, every paper in the next week's class will be assigned to one or two different students (or staff members). The student (or staff member) is expected to lead the discussion of the paper for about 30 minute, and then followed by (or interleaved with) Q&A (question-and-answer) session.

According to the size of the class, each student may present more than one papers during one semester.

II. Project and final report component ( 60%).

You are required to choose freely a recent paper from ICCV, ECCV or CVPR (since 2002).

(1) Implementation and improvement (10%)

You will be required to implement the algorithm proposed in the paper; It is not meant to be a complete implementation, but just as a proof-of-concept demonstration program.

You are encouraged to make any possible improvements to the paper ( this is optional). If doing so, you should highlight those of your own contributions in the final report. Any improvement will be awarded extra credit points (up to HD). You must be able to substantialize your contribution in your report, i.e., using convincing experimental results to justify that they are significant, and they are better.

You may choose to use Matlab , or C/C++, or else, as the programming language.

(2) Technique Report (30%)

Then write a 10~12 pages Technical-Report based on your own experiment; You report should be self-contained, should real like a conference paper, and should follow the format of a formal conference submission (using Springer LNCS single-side typesetting and Latex template).

(3) Final Presentation (20%)

Present your work at the final project-report class session. Each presentation is limited to 15~20 minutes;

Paper List (for class presentation/discussion):

1. Stefan Roth and Michael J. Black, Fields of Experts: A Framework for Learning Image Priors, CVPR

2. Grauman and T. Darrell. The Pyramid Match Kernel: Discriminative Classification with Sets of Image Features. ICCV

3. Categories S. Lazebnik, C. Schmid, and J. Ponce, Beyond Bags of Features: Spatial Pyramid Matching for Recognizing Natural Scene , CVPR

4. Y. Rubner, C. Tomasi, and L. J. Guibas, The earth mover's distance as a metric for image retrieval, IJCV.

5. A. Torralba, K. P. Murphy and W. T. Freeman, Contextual Models for Object Detection using Boosted Random Fields, NIPS.

6. Object detection using 2D spatial ordering constraints, CVPR.

7. D.Forthy, Detecting, locating and recovering animal in, CVPR

8. Amit Agrawal, Ramesh Raskar, and Rama Chellappa, What Is the Range of Surface Reconstructions from a Gradient Field? ECCV.
9. Brown et al, “Multi-Image Matching using Multi-Scale Oriented Patches”, CVPR 2005

10. Aharon Bar Hillel Tomer Hertz Daphna Weinshall, Efficient Learning of Relational Object Class Models, ICCV 2005

11. Anat Levin, Rob Fergus, Fredo Durand, William T. Freeman. Image and Depth from a Conventional Camera with a Coded Aperture. SIGGRAPH 2007.

12. Ido Omer, Michael Werman, Bottom-Up & Top-down Object Detection using Primal Sketch Features and Graphical Models., CVPR 2006.

13. A. C. Berg, T. L. Berg, J. Malik. Shape Matching and Object Recognition using Low Distortion Correspondence, CVPR 2005 .

14. Yakov Keselman et al, Many-to-Many Graph Matching via Metric Embedding, CVPR 2003.

15. A. Thayananthan,et al. Shape Context and Chamfer Matching in Cluttered Scenes , CVPR 2003.

16. Ido Omer. Michael Werman, The Bottleneck Geodesic: Computing Pixel Affinity, CVPR 2006.

17. Localization in urban environments: monocular vision compared to a differential GPS sensor, CVPR 2005.

18. Efficient image matching with distributions of local invariant features, CVPR 2005.

19. Category Recognition from Pair wise Interactions of Simple Features , CVPR 2007.

Course Announcement 2005-2006

*******************************

Supervised Literature Reading Coursework in Autonomous Systems
and Sensing Technology

Formal Description of course content
(to appear on Graduate Course Award Certificate) This is a paper
reading-and-presentation course. It is designed to broaden and
extend the knowledge base of participants in the most recent developments
of autonomous systems and intelligent sensing technology. The topics cover
computer vision and graphics, image processing, pattern recognition and
robotics.   It also serves the purpose of enhancing the students' scientific
communication and presentation skills, as well nurturing good research
habits.

Informal Description

Autonomous system and sensing technology is a challenging field with rapid
developing and tremendous promise.   With greatly improved computing
capacity, the promise of this modern field attracts much attention, yet the
challenges remain.   This course fulfils several important purposes, both
technically to offer students the opportunity to follow and grasp some of
the state-of-the-art achievements in this field, and practically to enhance
their academic communication skills.

At every session of the course ( a weekly 1.5-hour unit), each participant
is expected to engage in discussion of technical papers from recent world
major conferences
in computer vision, image processing, pattern recognition, graphics and
robotics etc.   A tentative list of these conferences includes ICCV,
ECCV, CVPR, SIGGRAPH, ACCV, ICPR, ICIP, ICRA, IROS etc. And the papers are
carefully chosen by the academics from RSISE/ASSeT.

Besides presenting one regular paper, each student who enrols in this course
is also required to present a talk on his research progress.   Having given
both presentations and passing the assessment, student will be accordingly
credited at the end of each course semester.

Coursework Coordinators:
Hongdong Li (RSISE, Information Engineering)

Prerequisites entry requirements:

This course will be given to RSISE and NICTA-endorsed graduate students but
registration is open to graduate students at ANU campus. Auditing of the
seminars by course participants will be available.

Assumed knowledge of course

Basic knowledge and skills in engineering, image processes, robotics and
computing.

Presenters

All participating students supervised by academic staff members from
RSISE/NICTA-ASSeT group.

Locations and Time

This course will be conducted in the video-conferencing lecture room, A207,
on the second floor of RSISE Building at ANU campus, at every Thursday
afternoon 3:30 PM ~ 5:00 PM.

Important date :

Start date:
End date: November 2005

Workload

For each student, the total study time is estimated 20 hours in class, and
40 hours after class.

Examiners

Two academics from RSISE/NICTA-ASSeT will review each student's
presentations. The following academics may be responsible for
reviewing any particular presentation. Note that other suitably
qualified academics who commence at ASSeT subsequently may also
review participant presentations:

Fees

There will be a $40 administration fee for formally enrolled students. For
RSISE students, this will be paid by RSISE. NICTA-endorsed students external
to RSISE should apply to their program leader for payment. There is no fee
for those auditing the course.

********************************

I am working towards a new course on modern pattern recognition technologies.

Prior to this, I have conducted the following courses/lectures:

1. Postgraduate students (1999,2000,2001,2002):

Modern Adaptive Signal Processing

-- Outline: model based spectrum estimation, DOA algorithm, adaptive filters, robust estimation, HOS, Blind source separation, and wavelet.

2. Undergraduate students: ( 1998-2001)

The Art of Software Programming.

Outlines: Data structure, algorithms and optimization, multi-task operating system, software engineering, database system, tcp/ip, Real-time software and RTOS.

3. Postgraduate students, ( 2002, 2003 )

Pattern Recognition, based on the books of Pattern Recognition, Second Edition by Sergios Theodoridis, Konstantinos Koutroumbas , and

Pattern Classification (2nd Edition) by Richard O. Duda, Peter E. Hart, David G. Stork