[ Last web update: May 8, 2013] Contact:milan-sonka@uiowa.edu

Department of Electrical & Computer Engineering
55:248 Advanced Image Processing

Spring 2013

Home

Course Information

Syllabus

Lecture Notes

Homework

News

Supplementary Materials

 

Homework - Projects:

Project 1: Camera rectification

  • due 2/19/13 midnight (email)
  • hand in printed report in class on 2/20/13 + be ready to give a brief presentation and demo on 2/20/13

Write a program that rectifies a pair of images from your digital camera.

  • Use your digital camera to acquire at least 3 pairs of stereo images. Use different complexity of mutual camera position/orientation starting from a simple one to a more general one.
  • Following the method and algorithm given in Section 11.5.6, write a program rectifying the stereo image pair - similar to situation on Fig. 11.12 of the textbook.
  • You may find useful Matlab functions and additional explanation given in the Matlab companion to the main textbook: Image Processing, Analysis & and Machine Vision - A MATLAB Companion (Paperback) - by Svoboda, Kybic, Hlavac (ISBN: 0495295957)
  • Web page http://visionbook.felk.cvut.cz/ includes code for many useful Matlab functions described in the companion.
  • You may also see and read Image rectification
  • You are not required to use the algorithm given in the book, and not required to use any of the code provided in the Companion book.
  • You can choose a programming environment of your choice.
  • Prepare a report as a PDF file with high-quality images and printout of the code and email to the instructor by deadline.

Project 2: Mathematical Morphology "light"

  • due 2/26/13 midnight (email)
  • hand in printed report in class on 2/27/13 + be ready to give a brief presentation and demo on 2/27/13
  • There will be no class on Monday 2/25/13 - use the time for working on this assignment

Using ultimate erosion, develop an approach for counting coffee beans (Fig. 13.5 of the Companion - Web page http://visionbook.felk.cvut.cz/ includes code for many useful Matlab functions described in the companion and the original images).

  • Test on the coffee beans image and demonstrate robustness (or lack of) with respect to the initial segmentation of the beans (binarization).
  • Apply the same approach to Fig. 13.7 of the Companion - can you determine the number of small and large scales of the gecko skin using ultimate erosion? Perhaps augmented by other morphologic approaches?
  • Use *only* mathematical morphology for this project.
  • You are not required to use the algorithm given in the book, and not required to use any of the code provided in the Companion book.
  • You can choose a programming environment of your choice.
  • Prepare a report as a PDF file with high-quality images and printout of the code and email to the instructor by deadline.

Project 3: Mathematical Morphology - Watershed

  • due 3/12/13 midnight (email)
  • hand in printed report in class on3/13/13 + be ready to give a brief presentation and demo on 3/13/13
  • There will be no class on Monday 3/4/13 and/or Wednesday 3/6/13- use the time for working on this assignment

Part A

  • Use the markers obtained by ultimate erosion (function ulterosion, Project 2, p. 184 of Companion) to segment the binary image of coffee beans using watershed transformation.
  • Web page http://visionbook.felk.cvut.cz/ includes code for many useful Matlab functions described in the companion and the original images
  • First compute the distance function of the binary image (use function :bwdist). Then impose the extracted markers on an inverted image of the distance function (use function :imimposemin). Finally, find watershed regions using the watershed
    transformation (use function :watershed).

Part B

  • Modify the watershed algorithm (function wshed, p. 189) so that it segments objects and borders separately. Apart from extracting object markers, it is necessary to extract markers for borders.
  • This can be done by inverting the object markers and by subsequent thinning (use function :imopen and function :bwmorph with parameter 'thin'). Do the watershed segmentation on a gradient image which has high intensity on borders and objects.
  • Use function :gradient to compute the gradient (see function mgvf, p. 92 of the Companion).

Note:

  • Use *only* mathematical morphology for this project.
  • You are not required to use the algorithm given in the book, and not required to use any of the code provided in the Companion book.
  • You can choose a programming environment of your choice.
  • Prepare a report as a PDF file with high-quality images and printout of the code and email to the instructor by deadline.

Project 4: Skeletonization

  • due 3/29/13 midnight (email)
  • hand in printed report in class on 4/1/13 + be ready to give a brief presentation and demo on 4/1/13

Part 1:

  • Read [Manzanera et al., 1999] paper (Manzanera A., Bernard T. M., Preteux F., and Longuet B. Ultra-fast skeleton based on isotropic fully parallel algorithm. In Proc. of Discrete Geometry for Computer Imagery, 1999.)
  • See
  • Write a program for skeletonization of binary objects using MB and MB2 algorithms
  • Apply on binary images (or binarized images) of your choice as well as to suitable images used in earlier projects.

Part 2:

  • Consider extension of this algorithm to 3D - can you design an algorithm/masks that would extend this work to find 3D skeletons? (Note that this may not be easy at all.)
  • Outline the main principles of your algorithm and demonstrate on a toy example that your approach works. (The goal is to show that your approach is applicable to at least a subclass of 3D objects.)
  • Can you find a counter-example = when your approach would fail? (Can you show that you are aware of "corner cases" that would require additional work?)
  • If you find time and interest, implement the 3D version of *your* algorithm extending one of the MB or MB2 algorithms.

3D Images

  • 3D brain image is available as TIFF and RAW, 256x256x129, 16 bits/pixel
  • threshold or segment as needed

Project 5: Registration

Teams:

  1. Victor/Ray
  2. Samantha/Andrew
  3. Saleh/Ali
  4. Emily/Kristin
  5. Farley/Zeng
  6. Teresa/Kyle
  7. Wenxiang/Eric
  8. Abhay/Jason
  • You will be working in pairs for this assignment - choose your partner and email me the "team" composition by Wed 4/24/13
  • PDF of complete draft due 5/6/13 midnight (email, one report per team) - see below for all required sections of the report
  • You will get comments back by 5/8/13 class time
  • Email your final PDF *and* your PPT slides to support your 10-min presentation by 5/13/13 midnight AND hand in printed report in class on 5/15/13 + be ready to give a 10' presentation and a brief demo on 5/15/13 starting at 2:30pm. We will do our best to be done by 5:30pm. We will meet in 3210 SC.

Required report formatting and sections:

  • The report should be no longer than 4 pages, double column format, just like a format required for a typical IEEE conference or journal.
  • References must be included within the 4 pages - not more than 1/2 column of references.
  • Use LaTeX or WORD - your choice --- see (http://www.ieee.org/conferences_events/conferences/publishing/templates.html)
  • Sections
    • Introduction - describe assignment, give an overview of approaches, cite relevant previous work
    • Methods - which methods you used, describe in sufficient detail
    • Experimental Methods - what data you used, how you approached assessment, what indices you used to determine how well your method worked.
    • Results - give your results according to what you used or assessment as described in the previous section
    • Discussion - how is what you did good/bad, compare with results of others, explain why it did better/same/worse
    • Conclusion
    • References
    • Any code you develop - please add as Appendix

Task:

  • Take 9 sets of MR/CT/PET images (0001-0009) from http://www.insight-journal.org/rire/ and perform cross-modality image registration (always pairwise)
    • choose whether you want to use PD, T1, or T2 images for MR modality - but you need to use at least one
    • this said, of course you can attempt using two or all three MR dataset types
  • Use the following methods:
    • Point-based rigid transformation - based on manually-identified landmarks
    • Point-based transformation with anisotropic scaling - based on manually-identified landmarks (scale your data to generate proper data set for this task)
    • Iterative closest point algorithm - apply to CT-MR and CT-PET registration. You will need to segment the skull surface from CT and identify landmarks on the skull surface manually (or using any other approach). Also, you will likely need to implement this approach yourself
    • Intensity-based using joint PDF
    • Intensity-based using mutual information - you may decide to use MI with or without normalization
    • You must program at least 2 methods yourself (each team member programs at least one approach); you can use ITK-based or other freely available code for the other methods (the complexity of approach you choose to code yourself will have effect on the number of points you get for the coding part of the project)
    • You may want to use 3D Slicer for visualizations. You may use registration algorithms included in Slicer for a max of 2 approaches.

Evaluation

  • In the evaluation, design the most relevant assessment criteria that will be applicable to your registration evaluation.
  • Compare performance of all implemented methods - use pairwise statistical assessment to determine whether performance of individual approaches is statistically significantly different from the other.
  • In Results, use checkerboard approach to show the registration outcomes, e.g., like here.
  • In Discussion, consider accuracy of your techniques and determine for what applications would the implemented registrations be sufficiently accurate.

=======