### IBMR Assignment 1

```Stitching Photo Mosaics
•
•
•
•
Stitching photos to construct a wild-view scene.
Part1: CORNER DETECTION
Part2: PERSPECTIVE MAPPING and MOSAICING
Handout after Part2 Finished
PERSPECTIVE MAPPING and
MOSAICING
• Read n>2 images, and create an image mosaic by registering,
projective warping, resampling, and compositing them.
• (bonus) multiband blending, SIFT ,panorama or other methods
mentioned in class.
•
•
•
•
•
Shoot the Pictures
Recover Homographies
Warp the Images/ Image Rectification
Gain Compensation
Blend the images into a mosaic
• You may use the photos on the webpage, but shoot your own
photos and mosaic them will get bonus credit.
• Shoot photos as:
• Overlap the fields of view significantly. 40% to 70% overlap is
recommended.
• Construct a linear system as: p’=Hp, where p’ and p are
correspondence points.
• Follow the Lecture 8 page 6~9. You may try Affine
mappings(DOF=6) or Projective mappings(DOF=8).
• Solve Ax=0
• Source scanning(forward mapping) or destination
scanning(inverse mapping).
• You will need to avoid aliasing when resampling the image.
• Be careful of the size of the resulting image.
• Find the optimize gains of gi according to means of
overlapping regions between image pair i and j.
• Linear blending by the weights:
where w(x) varies linearly
from 1 at the centre of the
image to 0 at the edge.
A
B
(, ) =
• Multi-band blending (bonus):
∗  −  +  ∗ ( − )
−  + ( − )
•
Band 1 scale 0 to σ
•
Band 2 scale σ to 2σ
•
Band 3 lower than 2σ
• Your own project1a code.
• A C called matlab library.
• to calculate inverse matrix , SVD or etc.
• Basic: 75%
•
•
•
•
Harris Corner Detection + KNN (Hw1a)
RANSAC
Projection Mapping / Affine Mapping
Image Warping
• Bonus:
•
•
•
•
•
•
•
•
Non-Maximum Suppression
KD Tree
SIFT
Gain Compensation
Linear Blending
Multi Blending
Stitching your own photos
Others
5%
5%
15%
10%
5%
10%
5%
• 11/22 11:59:59pm
• Upload your program & report to:
•
•
•
•
host : caig.cs.nctu.edu.tw
port : 30021