CS/ECE 545 Projects

The group projects will develop software to solve typical image processing problems.

Each group will analyze a problem or class or problems, explore algorithmic solutions, write software to address the problem, write a brief summary of what was learned, and give an oral presentation to the class at the end of the course.

Materials that will be made available include bibliographies and sets of typical data.

Team Size
Teams should have 3-4 students. If you are considering a different size team, check with the instructor.

What to Submit
At the final day of class, turn in a project report explaining the problem, your approach, and assumptions, and tradeoffs. Show a convincing set of results and explain how your algorithms worked and why they didn't. We are more interested in seeing that you had a reasonable approach and can explain what happened than that you got the best possible solution. Use correct grammar and spelling. Include your programs.

Prepare a 10-minute presentation of your project for the final day of class. Everyone in your group should have an opportunity to explain part of the project. Use PowerPoint or Acrobat or equivalent for your presentation. We recommend against showing your complete programs.

Project 1 Tomographic Reconstruction

This project is about how to reconstruct 3D visual data from collections of tomographic data. A typical example is the medical Computer Aided Tomograph (Cat Scan). X-ray images are taken through the body from many angles around the central axis. Each image shows the total x-ray absorption in some direction. From a suite of such images, a 3D model of the body can be reconstructed.

Some of the issues to be addressed in this project are how to reconstruct 3D data from 2D projection, and how to deal with incomplete data (such as a missing image in the data set).

Dataset is in tomo/

Project 2 Missing Data

This project is about how to synthesize realistic data to replace missing data. For example, sensors occasionally have defects that result in entire rows or columns of missing data. The goal is to develop algorithms that can supply reasonable replacement data such that the missing data is not obvious. Consider using frequency-domain as well as, or in conjunction with, spatial domain techniques.

Datasets are in missing/

Project 3 Combining Multiple Images

Often times images are reconstructed from a collection of smaller images. For example, overlapping areal photographs are often combined to create a mosaic image of a larger area. Also, multiple images of the same object made with small shifts of the camera, can be combined to produce a composite image with greater resolution that any of the individual images. This project will explore the problems of combining multiple images.

Datasets are in superres/

Project 4 Rotation

Sometimes, it is necessary to align images by rotation and shifting. This project requires estimation of the rotation between image pairs and correction for the rotation. Also, 4 images of another scene are supplied with different rotations. THese must be combined into a single image.

Datasets are in rotation/

Project 5 Shredder Challenge