EEGR6118 Applied Digital Image Processing (Fall 2008)
Instructor: Prof. T. Weldon
Course Outline
Course overview
Overall, the course philosophy will be "to learn by doing."
Students who have taken ECGR 4103/5103 Applied Computer Graphics will be best prepared for the course, otherwise students should have a good background in Fourier transforms or digital signal processing. We hope to set aside several weeks at the end of semester to allow students to independently pursue a topic of interest. You may choose Java/NetBeans/C++ or Mathcad to develop your project code to implement algorithms. This is a good opportunity to add programming language skills to your resume.
Some links of general interest:
For a quick tutorial on Java and Netbeans. see:
NetBeans and Java tutorial
For more details on getting Java/Netbeans for your home PC, see:
http://www.netbeans.org/
For a quick tutorial on Mathcad, see:
Mathcad Introduction
For a quick tutorial on C++, see:
Intro to C++, UNIX, and make
vecdemo:
a simple C++ program example.
Projects
Images files for use in the course:
Images
If you choose to program in Java/NetBeans/C++, you may wish to use ImgPro-jv41 as the starting point for your projects.
ImgPro-jv41
is a Java/Netbeans-4.1 program for MSwindows that reads ".gif" and ".jpg" files, has a built-in logfft function, and writes ".png" output files.
Project 1: Transforms
Warning: these files may not download correctly using MS Intrnet Exxplorer,
Use Firefox to download files (right-click, save-link-as).
pdf file
or
Mathcad 13 mcd file
A mathcad demo showing a more elaborate centered Fourier display is
imageFFT-12aug06.zip
or in
pdf form.
Project 2: Subband/Wavelet representations
pdf file
or
Mathcad 14 xmcd file
Reference papers available on
IEEE xplore
:
Quadrature mirror filter banks, M-band extensions and perfect-reconstruction techniques" P. Vaidyanathan, IEEE ASSP Magazine, [now IEEE Signal Processing Magazine] Volume 4, Issue 3, Part 1, Jul 1987 Page(s):4 - 20.
Wavelets and filter banks: theory and design M. Vetterli and C. Herley, IEEE Transactions on Signal Processing, Volume 40, Issue 9, Sept. 1992 Page(s):2207 - 2232.
Image coding using wavelet transform Antonini, M.; Barlaud, M.; Mathieu, P.; Daubechies, I.; Image Processing, IEEE Transactions on Volume 1, Issue 2, April 1992 Page(s):205 - 220.
Project 3: Image Enhancement
mean filter, median filter, sharpening, histogram equalization
pdf file
or
Mathcad 2000 mcd file
Project 4: Morphology
dilation, erosion, open, close
pdf file
or
Mathcad 2000 mcd file
Project 5: Edge Detection
Laplacian, Sobel, and Hough transform
pdf file
or
Mathcad 2000 mcd file
Project 6: Feature Selection and Texture
Laws features, Gabor filters, B-distance
pdf file
or
Mathcad 2000 mcd file
Project 7: Segmentation
Features, Bayesian classifier, nearest-neighbor classifier
pdf file
or
Mathcad 13 xmcd file
or (do not use Mathcad 2000 version for class!)
older Mathcad 2000 mcd file
Project 8: 3D
pdf file
or
Mathcad 2000 mcd file
Old versions of project files.
Warning: Do not use the following versions for class!
Project 1:
old Mathcad file
Project 2:
old Mathcad file
Final Project
You may propose a special project only if you get approval to do so beforehand (this is not a guarantee that your proposal will be approved).
No segmentations are allowed using color images, all projects must use grayscale images, except for pre-approved special projects.
Your final project must be a complete image segmentation system that includes as a minimum:
An example image that is to be segmented
3 features (i.e., Gabor, Laws, Sobel, etc.)
A vector classifier (Bayesian, nearest neighbor, etc.)
Either: 3 classes with morphological post-processing, or more than 3 classes. At least one class must be textured, unless you have some special project of equal complexity with prior approval.
Some ideas for projects:
See the
Images
page for ideas
See the images on the CD in the textbook
See the textbook
Segment the satellite image into rural, urban, ocean classes
Segment a brain scan into bone, brain, blood vessel
Photograph an outdoor scene and segment image into 4 classes: cars, people, buildings, other
Given a sequence of 2 or 3 photos at increments of 10 meters, estimate the ranges, and generate a 3D image (gamestudio mdl format file) of Woodward Hall
In a sequence of images of a moving car, identify the license plate, car, and letters
Proposal (10 percent of Final Project Grade)
Due Nov. 10.
4 pages maximum.
Prepare a proposal for your final project.
Your proposal must be in the following format:
Use this template:
Proposal Template
Page 1: cover page with proposal title and name
Page 2: Executive summary:
describe your project in one half of a page or less, and on the bottom half of the page include a picure of an example image to be segmented, and a sketch of what you want the final segmented image to look like. Caption and label the image (i.e., Fig. 1. Image showing 3 classes to be segmented: trees, roads, and buildings.
Page 3: System Block diagram.
At the top of the page provide a block diagram including separate blocks for each feature (Laws1, Laws2, Gabor 1, etc.), a block for the classifier, and a block for postprocesing. In the blocks, indicate the type of classifier/postprocessing. On the rest of the page, include the following headings and a paragraph describing each:
Features
Classifier Design
Post-processing
Page 4: Test data
At the top of the page provide one or more images showing all of the classes that you want to segment. On the remainder of the page, describe the image classes (trees, roads, bone, muscle, chairs, etc), data source (where images came from, website, etc.), and segmentation problem (segment trees from roads). Include the following headings and a paragraph describing each:
Image classes
Data source
Segmentation problem
Design Review (20 percent of Final Project Grade)
Due Nov. 24.
Use this template:
Design Review Template
You must demonstrate your feature images in class and turn in report. The features should clearly show that you can segment the image. It would be wise to show your classifier, too.
Final Presentation (20 percent of Final Project Grade)
Due Last day of class (Dec 8).
(20 percent of Final Project Grade)
You must demonstrate your final project in class. Demonstrate that you can automatically segment the image.
Final Presentation make-up: If your final presentation is not accepted, you have opportunity to do a second presentation, but with a 25% reduced grade. All make-up demonstrations will be conducted during the regularly scheduled final exam period, at which time all final reports are also due.
Final Report (50 percent of Final Project Grade)
Due before end of the regularly scheduled final exam period.
10 pages maximum.
All Final Project reports AND source code must be emailed.
Your Final Report must be in the following format:
Use this template:
Final Report Template
Page 1: cover page with proposal title and name
Page 2: Executive summary:
describe your project in 0.5 pages or less, and on the bottom half of the page include a picure of an example image to be segmented next to a picture of the segmented image. The test image must contain all of the classes that you intend to segment. Caption and label the image (Fig. 1. (a) Image showing classes to be segmented: trees, roads, and buildings. (b) Segmented image)
Pages 3-4: System Block diagram.
At the top of the page provide a block diagram including separate blocks for each feature (Laws1, Laws2, Gabor 1, etc.), a block for the classifier, and a block for postprocesing. In the blocks, indicate the type of classifier/postprocessing. On the rest of pages 3-4, include the following headings and a section describing each block in detail:
Features (masks used, or filter frequencies and bandwidth, etc.)
Classifier Design (Bayes, neares-neighbor, etc.)
Post-processing
Pages 5-6: Test data
At the top of the page provide one or more images showing all of the classes that you want to segment. On the remainder of the pages 5-6, describe the image classes (trees, roads, etc), data sources (where images came from), and segmentation problem (segment trees from roads). Include the following headings and a section describing each in detail:
Image classes
Data source
Segmentation problem
Pages 7-8: Feature data
At the top of the page provide one or more images showing a test image along with the corresponding feature images for all of your features. The test image must contain all of the classes that you intend to segment. On the remainder of the pages 7-8, describe the feature images. Include the following headings and a section describing each in detail:
Feature images
Page 9: Segmentation Results
At the top of the page provide one or more images showing pairs of images, with the test image on the left and segmentation results on the right. The test image must contain all of the classes that you intend to segment. Include the following headings and a section describing each in detail:
Segmentation results (description of segmented images)
Measured segmentation error rate
Page 10: Conclusion
All Final Project reports AND source code must be emailed.
Project Problem Resolution
Guidelines
(If members of a group are not contributing)
Announcements
Misc
Project core:
The projects will be based on a core image viewer with extra buttons to add new functions for your projects.
Go to project core page
Images:
A number of images (256x256 pixel) are used for the projects.
Images
Other information
old exams
Spring 1998
Old Projects
Project 1
Project 2
Project 3
Other information
Project Problem Resolution
Guidelines
(If members of a group are not contributing)
Copyright information
Copyright © 1995-2012 by Thomas Weldon, all rights reserved.