-
Notifications
You must be signed in to change notification settings - Fork 0
/
README
25 lines (22 loc) · 837 Bytes
/
README
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
# Old strategy
1. Detect and extract features (ORB or SIFT)
2. Matched features between images
3. Found patches around the matches
4. Used those patches to estimate an exposure correction for each pixel (least squares)
5. Took mode (bin) of the least squares parameters
6. Used that exposure correction on whole image
# New strategy
*. Detect and extract features (ORB or SIFT)
*. Matched features between images
3. Estimate projective transform
4. Sample shared pixels from projective transform
5. Use those samples to estimate tempurature, tint, and lightness transforms
6. Transform the image using those transforms
# Project proposal
Embed GPU program in python bindings
Steps 1-3 are in a calibration procedure
Steps 4-6 of New strategy done on GPU
# Next steps
- working prototype of New strategy
- write proposal
- implement it