This homework covers several standard stereo methods: local using windows, scan-line using DP, and global using graph cuts. The overal goal is to understand the challanges of stereo and learn different algorithms for multi-label optimization problems.
There is no need for any user interactions and I do not provide any initial project. If you like, you can start from the project provided for the previous assignment since you will
need to load/visualize/save images and to use max-flow library available in HW2. Calibrated stereo pairs are provided at the "image samples" section of the course web site. For each pair
you need to estimate the range of disparities: integers in some range from 0 to d_max. You can choose either left of right image as your base image for each pair, but make sure you apply horizontal shifts (disparities) in the right direction.
Window-based stereo: Implement window-based stereo and compare the results for windows of different sizes.
Scan-line stereo: Implement scan-line stereo using viterbi algorithm (DP) to optimize the sum of photo-consistensies and smoothness along each scan line. Compare three smoothness terms penalizing quadratic, absolute, and "truncated" absolute difference between neighboring pixels. Also, try two versions with and without "static cues". Bonus points: compare to shortest path (Dijkstra) method for scan-line stereo presented in class.
Global stereo: Implement global stereo using absolute-difference (or TV) smoothness for the disparity map. Use Ishikawa graph cut consruction. Note that additional slides showing how to avoid folding were added to topic 7. Compare two versions with and without "static cues".
MINI PROJECT FOR GRAD STUDENTS (extra credit for undergrads). Implement global stereo with trancated linear smoothness term using a-expansion and compare to the global stereo results for TV (as above). Note that this mini-project could be submitted as a separate report (pdf file) at a later date (a separate deadline is posted).
What to submit
Submit a pdf file with your report (no more than 1.5 pages of text, but any number of images representing your results) summarizing your efforts in achieving the goals described above.
In general, images of your resulst are highly encourages as they are an ideal way to represent your experiemnts in image analysis. The report should be submitted electronically via OWL. You should also submit your code (in a .zip file). The project report and code should be your independent effort.