ECE 278a

Topics in Digital Image Processing

11/29: Registration example images for your final homework are now available. download the zip file (about 5MB).

helpful links:

forward mapping: http://blogs.mathworks.com/steve/2006/04/28/spatial-transforms-forward-mapping/

inverse mapping: http://blogs.mathworks.com/steve/2006/04/28/spatial-transforms-inverse-mapping/

11/29: Image Restoration slides posted (also slides on histogram equalization).

11/25 New notes on Registration, Slides on Registration and the final home work on registration (matlab programming assignment) are posted/ H/W is due on the last day of classes, Friday Dec 7, by 5pm.

11/21 More segmentation notes (active contours and snakes) posted. Solutions to homeworks.

11/21: Final examination will be a take home exam. You can pick up the exam on Monday, Dec 10, between 11am-12 noon. The answers are due before 5pm on Tuesday, Dec 11. I will have my office hours from 5-6 pm on Monday and 11-12noon on Tuesday.

11/19 Notes on Anisotropic diffusion posted

11/12 HW #6 Registration programming assignment due Nov 19. You can now download the solution to the first MATLAB assignment that you may use for this h/w. See below (in HW2).

11/06 Slides and Notes on segmentation are posted.

11/05 HW #5 Wavelet Programming Assignment Due Nov 14.

10/31 Registration Slides #2 posted. Also, the registration notes updated.

10/29 Registration Slides #1 posted (see below)

10/24: Corrections to HW#3.

Q4: note that u_s and \Delta x are related: u_s = 1/(\Delta x). similarly, v_s and \Delta y.

Q6: the delta term inside the summation is missing the summation indices k and l; it should be \delta(u - k u_0, v - l v_0).

Instructor: Manjunath (Office Hours: T 11-1150am and Th 1-150pm, or by appointment)
Teaching Assistant: Luca Bertelli (Office Hours: M 9-10am and F 4-5pm)

PS: You may contact by e-mail: manj at ece.ucsb.edu, but please make sure that you clearly note the subject heading as ECE 278a and purpose of the email at the very beginning. I get quite a few emails and will be difficult to respond in a timely manner otherwise.

Topics

Introduction: Digital picture transforms, 2-D DFT; Random fields and statistical image models. Image representations: 2-D sampling theory, generalization to random fields, representation using orthonormal basis functions.
Digital Image Transforms and Coding: Overview of transform compression; Various transforms--Karhunen-Loeve, Fourier, Cosine, Hadamard, Walsh, 2-D Wavelet; Still image coding standards-JPEG and JPEG-2000; Video coding. Reading: Compression slides from ece178 part1, part2. intro slides on wavelets.
Image Restoration: Inverse and Least-squares filtering. Slides (this is a large, 28MB file).
You can also download the histogram equalization discussion slides here.
Image Reconstruction from Projections: Image Projections, Radon transform and the Fourier Slice Theorem; Filtered back projection algorithm.
Image Registration: Feature selection, feature correspondence, estimating homographies (and RANSAC) (guest lectures by Dr. Marco Zuliani).
Slides#1 (link removed)
Slides #2 (full set, includes first set)
Updated Notes. Also See reading list below.
Final Notes (posted 11/25)
New Slides (posted 11/25)
Image Segmentation: edge based and region based methods. (guest lectures by Luca Bertelli).
Notes
Slides #1
Anisotropic diffusion notes Snakes Active contours without edges

Grading: 60% for H/W, including MATLAB assignments; 40% for the final. Homeworks are due in class. Late h/w will not be graded.

References

No required text; however, if this is your first introduction to image processing, I recommend the book by Gonzalez and Woods (Digital Image Processing, 2nd Ed., Prentice Hall, 2002) that I used in the undergraduate ECE 178. The book by Anil Jain (Fundamentals of DIP, Prentice Hall 1989) is quite comprehensive (a good reference book to have) though a bit outdated.

Additional Reading:

Image Registration Papers:

Image Registration survey by Zitova and Flusser;
Condition theory based point matching by Zuliani et. al.;
SIFT features by Lowe;
Panaromic image stitching by Brown and Lowe;
RANSAC algorithm by Fischler and Bolles.

Homework #1: Due Oct 8, 2007 in class. solution

Homework #2: Due Oct 15, 2007 in class. Read Chapter 2 and work out the MATLAB exercise. References will be periodically updated.
New: http://vision.ece.ucsb.edu/~zuliani/Code/Matlab/compute_image_derivatives.m