photo stitching panoramas from multiple images
DESCRIPTION
04/05/11. Photo Stitching Panoramas from Multiple Images. Computer Vision CS 543 / ECE 549 University of Illinois Derek Hoiem. So far, we’ve looked at what can be done with one image. Recover basic geometry using vanishing points Find image boundaries and segment objects - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/1.jpg)
Photo StitchingPanoramas from Multiple Images
Computer VisionCS 543 / ECE 549
University of Illinois
Derek Hoiem
04/05/11
![Page 2: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/2.jpg)
So far, we’ve looked at what can be done with one image
• Recover basic geometry using vanishing points
• Find image boundaries and segment objects
• Categorize images
• Find specific objects and detect objects that are part of some category
![Page 3: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/3.jpg)
What can we get from multiple images?
![Page 4: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/4.jpg)
What can we get from multiple images?
• Bigger, Better, Brighter, Sharper images– Panoramas– Increased dynamic range– Super-resolution– Reduced noise/blur
Product example: http://www.vreveal.com/
![Page 5: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/5.jpg)
What can we get from multiple images?
• Bigger, Better, Brighter, Sharper images– Panoramas– Increased dynamic range– Super-resolution– Reduced noise/blur
Product example: http://www.vreveal.com/
today
![Page 6: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/6.jpg)
What can we get from multiple images?
• Motion– Tracking– Optical flow– Action/activity recognition
Optical flow (source: http://www.borisfx.com/avid/bccavx/classic_features.php)
Tracking (from Deva Ramanan)
![Page 7: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/7.jpg)
What can we get from multiple images?
• Motion– Tracking– Optical flow– Action/activity recognition
Optical flow (source: http://www.borisfx.com/avid/bccavx/classic_features.php)
Tracking (from Deva Ramanan)
Thursday
![Page 8: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/8.jpg)
What can we get from multiple images?
• Motion– Tracking– Optical flow– Action/activity recognition
Optical flow (source: http://www.borisfx.com/avid/bccavx/classic_features.php)
Tracking (from Deva Ramanan)
April 19
![Page 9: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/9.jpg)
What can we get from multiple images?
• Motion– Tracking– Optical flow– Action/activity recognition
Optical flow (source: http://www.borisfx.com/avid/bccavx/classic_features.php)
Tracking (from Deva Ramanan)
April 21
![Page 10: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/10.jpg)
What can we get from multiple images?
• Depth and 3D structure– Two-view stereo– Multi-view stereo– Shape carving– Structure from motion
Next week
![Page 11: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/11.jpg)
Key ideas
• Correspondence– Detect interest points– Match two patches
• Alignment and model fitting– Solve for motion or depth or camera movement
from multiple matched points
![Page 12: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/12.jpg)
Today: Image Stitching• Combine two or more overlapping images to
make one larger image
Add example
Slide credit: Vaibhav Vaish
![Page 13: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/13.jpg)
Example
Camera Center
![Page 14: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/14.jpg)
Review of projective geometry
![Page 15: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/15.jpg)
Problem set-up• x = K [R t] X• x' = K' [R' t'] X• t=t'=0
• x'=Hx where H = K' R' R-1 K-1
• Typically only R and f will change (4 parameters), but, in general, H has 8 parameters
f f'
.
x
x'
X
![Page 16: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/16.jpg)
Image Stitching Algorithm Overview
1. Detect keypoints2. Match keypoints3. Estimate homography with four matched
keypoints (using RANSAC)4. Combine images
![Page 17: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/17.jpg)
Computing homography
Assume we have four matched points: How do we compute homography H?
Direct Linear Transformation (DLT)
Hxx '
0h
vvvvuvu
uuvuuvu
1000
0001
'
''
''
'
w
vw
uw
x
987
654
321
hhh
hhh
hhh
H
9
8
7
6
5
4
3
2
1
h
h
h
h
h
h
h
h
h
h
![Page 18: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/18.jpg)
Computing homography
Direct Linear Transform
• Apply SVD: UDVT = A• h = Vsmallest (column of V corr. to smallest singular value)
987
654
321
9
2
1
hhh
hhh
hhh
h
h
h
Hh
0Ah0h
nnnnnnn vvvvuvu
vvvvuvu
uuvuuvu
1000
1000
0001
1111111
1111111
Matlab [U, S, V] = svd(A);h = V(:, end);
Explanations of SVD and solving homogeneous linear systems
![Page 19: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/19.jpg)
Computing homography• Assume we have four matched points: How do
we compute homography H?
Normalized DLT1. Normalize coordinates for each image
a) Translate for zero meanb) Scale so that average distance to origin is sqrt(2)
– This makes problem better behaved numerically (see HZ p. 107-108)
2. Compute using DLT in normalized coordinates3. Unnormalize:
Txx ~ xTx ~
THTH~1
ii Hxx
H~
![Page 20: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/20.jpg)
Computing homography
• Assume we have matched points with outliers: How do we compute homography H?
Automatic Homography Estimation with RANSAC1.Choose number of samples N
HZ Tutorial ‘99
![Page 21: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/21.jpg)
Computing homography• Assume we have matched points with outliers: How do
we compute homography H?
Automatic Homography Estimation with RANSAC1. Choose number of samples N2. Choose 4 random potential matches3. Compute H using normalized DLT4. Project points from x to x’ for each potentially matching
pair:5. Count points with projected distance < t
– E.g., t = 5 pixels6. Repeat steps 2-5 N times
– Choose H with most inliers
HZ Tutorial ‘99
ii Hxx
![Page 22: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/22.jpg)
Automatic Image Stitching
1. Compute interest points on each image
2. Find candidate matches
3. Estimate homography H using matched points and RANSAC with normalized DLT
4. Transform second image and blend the two images
– Matlab: maketform, imtransform
![Page 23: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/23.jpg)
RANSAC for Homography
Initial Matched Points
![Page 24: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/24.jpg)
RANSAC for Homography
Final Matched Points
![Page 25: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/25.jpg)
Verification
![Page 26: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/26.jpg)
RANSAC for Homography
![Page 27: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/27.jpg)
Choosing a Projection Surface
Many to choose: planar, cylindrical, spherical, cubic, etc.
![Page 28: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/28.jpg)
Planar vs. Cylindrical Projection
Planar
Photos by Russ Hewett
![Page 29: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/29.jpg)
Planar vs. Cylindrical Projection
Planar
![Page 30: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/30.jpg)
Planar vs. Cylindrical Projection
Cylindrical
![Page 31: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/31.jpg)
Planar vs. Cylindrical Projection
Cylindrical
![Page 32: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/32.jpg)
Planar
Cylindrical
![Page 33: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/33.jpg)
Simple gain adjustment
![Page 34: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/34.jpg)
Recognizing Panoramas
Brown and Lowe 2003, 2007Some of following material from Brown and Lowe 2003 talk
![Page 35: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/35.jpg)
Recognizing Panoramas
Input: N images1.Extract SIFT points, descriptors from all
images2.Find K-nearest neighbors for each point (K=4)3.For each image
a) Select M candidate matching images by counting matched keypoints (m=6)
b) Solve homography Hij for each matched image
![Page 36: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/36.jpg)
Recognizing Panoramas
Input: N images1.Extract SIFT points, descriptors from all
images2.Find K-nearest neighbors for each point (K=4)3.For each image
a) Select M candidate matching images by counting matched keypoints (m=6)
b) Solve homography Hij for each matched image
c) Decide if match is valid (ni > 8 + 0.3 nf )
# inliers # keypoints in overlapping area
![Page 37: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/37.jpg)
Recognizing Panoramas (cont.)
(now we have matched pairs of images)4.Find connected components
![Page 38: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/38.jpg)
Finding the panoramas
![Page 39: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/39.jpg)
Finding the panoramas
![Page 40: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/40.jpg)
Recognizing Panoramas (cont.)
(now we have matched pairs of images)4.Find connected components5.For each connected component
a) Perform bundle adjustment to solve for rotation (θ1, θ2, θ3) and focal length f of all cameras
b) Project to a surface (plane, cylinder, or sphere)c) Render with multiband blending
![Page 41: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/41.jpg)
Bundle adjustment for stitching• Non-linear minimization of re-projection error
• where H = K’ R’ R-1 K-1
• Solve non-linear least squares (Levenberg-Marquardt algorithm)– See paper for details
)ˆ,(1 N M
j k
i
disterror xx
Hxx ˆ
![Page 42: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/42.jpg)
Bundle Adjustment• New images initialised with rotation, focal
length of best matching image
![Page 43: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/43.jpg)
Bundle Adjustment• New images initialised with rotation, focal
length of best matching image
![Page 44: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/44.jpg)
Blending
• Gain compensation: minimize intensity difference of overlapping pixels
• Blending– Pixels near center of image get more weight– Multiband blending to prevent blurring
![Page 45: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/45.jpg)
Multi-band Blending• Burt & Adelson 1983
– Blend frequency bands over range
![Page 46: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/46.jpg)
Multiband blending
![Page 47: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/47.jpg)
Blending comparison (IJCV 2007)
![Page 48: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/48.jpg)
Blending Comparison
![Page 49: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/49.jpg)
Further reading• DLT algorithm: HZ p. 91 (alg 4.2), p. 585• Normalization: HZ p. 107-109 (alg 4.2)• RANSAC: HZ Sec 4.7, p. 123, alg 4.6
• Recognising Panoramas: Brown and Lowe, IJCV 2007 (also bundle adjustment)
![Page 50: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/50.jpg)
Things to remember• Homography relates rotating cameras
• Recover homography using RANSAC and normalized DLT
• Bundle adjustment minimizes reprojection error for set of related images
• Details to make it look nice (e.g., blending)
![Page 51: Photo Stitching Panoramas from Multiple Images](https://reader035.vdocuments.mx/reader035/viewer/2022081519/568139c0550346895da16680/html5/thumbnails/51.jpg)
Next class
• Corner tracking and optical flow