10/19/10
DESCRIPTION
10/19/10. Image Stitching. Computational Photography Derek Hoiem, University of Illinois. Photos by Russ Hewett. Last Class: Keypoint Matching. 1. Find a set of distinctive key- points. 2. Define a region around each keypoint. A 1. B 3. - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/1.jpg)
10/19/10
Image Stitching
Computational PhotographyDerek Hoiem, University of Illinois
Photos by Russ Hewett
![Page 2: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/2.jpg)
Last Class: Keypoint Matching
K. Grauman, B. Leibe
Af Bf
B1
B2
B3A1
A2 A3
Tffd BA ),(
1. Find a set of distinctive key- points
3. Extract and normalize the region content
2. Define a region around each keypoint
4. Compute a local descriptor from the normalized region
5. Match local descriptors
![Page 3: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/3.jpg)
Last Class: Summary
• Keypoint detection: repeatable and distinctive– Corners, blobs– Harris, DoG
• Descriptors: robust and selective– SIFT: spatial histograms of gradient
orientation
![Page 4: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/4.jpg)
Today: Image Stitching• Combine two or more overlapping images to
make one larger image
Add example
Slide credit: Vaibhav Vaish
![Page 5: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/5.jpg)
Panoramic Imaging
• Higher resolution photographs, stitched from multiple images
• Capture scenes that cannot be captured in one frame
• Cheaply and easily achieve effects that used to cost a lot of money
• Like HDR and Focus Stacking, use computational methods to go beyond the physical limitations of the camera
![Page 6: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/6.jpg)
Photo: Russell J. Hewett
Pike’s Peak Highway, CO
Nikon D70s, Tokina 12-24mm @ 16mm, f/22, 1/40s
![Page 7: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/7.jpg)
Photo: Russell J. Hewett
Pike’s Peak Highway, CO
(See Photo On Web)
![Page 8: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/8.jpg)
Software
• Panorama Factory (not free)
• Hugin (free)
• (But you are going to write your own, right?)
![Page 9: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/9.jpg)
Photo: Russell J. Hewett
360 Degrees, Tripod Leveled
Nikon D70, Tokina 12-24mm @ 12mm, f/8, 1/125s
![Page 10: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/10.jpg)
Photo: Russell J. Hewett
Howth, Ireland
(See Photo On Web)
![Page 11: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/11.jpg)
Capturing Panoramic Images
• Tripod vs Handheld• Help from modern cameras• Leveling tripod• Gigapan• Or wing it
• Exposure• Consistent exposure between frames• Gives smooth transitions• Manual exposure• Makes consistent exposure of dynamic scenes easier• But scenes don’t have constant intensity everywhere
• Caution• Distortion in lens (Pin Cushion, Barrel, and Fisheye)• Polarizing filters• Sharpness in image edge / overlap region
• Image Sequence• Requires a reasonable amount of overlap (at least 15-30%)• Enough to overcome lens distortion
![Page 12: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/12.jpg)
Photo: Russell J. Hewett
Handheld Camera
Nikon D70s, Nikon 18-70mm @ 70mm, f/6.3, 1/200s
![Page 13: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/13.jpg)
Photo: Russell J. Hewett
Handheld Camera
![Page 14: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/14.jpg)
Photo: Russell J. Hewett
Les Diablerets, Switzerland
(See Photo On Web)
![Page 15: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/15.jpg)
Photo: Russell J. Hewett & Bowen Lee
Macro
Nikon D70s, Tamron 90mm Micro @ 90mm, f/10, 15s
![Page 16: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/16.jpg)
Photo: Russell J. Hewett & Bowen Lee
Side of Laptop
![Page 17: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/17.jpg)
Considerations For Stitching
• Variable intensity across the total scene
• Variable intensity and contrast between frames
• Lens distortion• Pin Cushion, Barrel, and Fisheye• Profile your lens at the chosen focal length (read from EXIF)• Or get a profile from LensFun
• Dynamics/Motion in the scene• Causes ghosting• Once images are aligned, simply choose from one or the other
• Misalignment• Also causes ghosting• Pick better control points
• Visually pleasing result• Super wide panoramas are not always ‘pleasant’ to look at• Crop to golden ratio, 10:3, or something else visually pleasing
![Page 18: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/18.jpg)
Photo: Russell J. Hewett
Ghosting and Variable Intensity
Nikon D70s, Tokina 12-24mm @ 12mm, f/8, 1/400s
![Page 19: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/19.jpg)
Photo: Russell J. Hewett
![Page 20: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/20.jpg)
Photo: Bowen Lee
Ghosting From Motion
Nikon e4100 P&S
![Page 21: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/21.jpg)
Photo: Russell J. Hewett Nikon D70, Nikon 70-210mm @ 135mm, f/11, 1/320s
Motion Between Frames
![Page 22: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/22.jpg)
Photo: Russell J. Hewett
![Page 23: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/23.jpg)
Photo: Russell J. Hewett
Gibson City, IL
(See Photo On Web)
![Page 24: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/24.jpg)
Photo: Russell J. Hewett
Mount Blanca, CO
Nikon D70s, Tokina 12-24mm @ 12mm, f/22, 1/50s
![Page 25: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/25.jpg)
Photo: Russell J. Hewett
Mount Blanca, CO
(See Photo On Web)
![Page 26: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/26.jpg)
Problem basics• Do on board
![Page 27: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/27.jpg)
Basic problem • 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 28: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/28.jpg)
Views from rotating camera
Camera Center
![Page 29: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/29.jpg)
Image Stitching Algorithm Overview
1. Detect keypoints2. Match keypoints3. Estimate homography with four matched
keypoints (using RANSAC)4. Project onto a surface and blend
![Page 30: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/30.jpg)
Image Stitching Algorithm Overview
1. Detect keypoints (e.g., SIFT)2. Match keypoints (most similar features,
compared to 2nd most similar)
![Page 31: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/31.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 32: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/32.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);
![Page 33: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/33.jpg)
Computing homographyAssume 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 Hartley and Zisserman p. 107-108)
2. Compute H using DLT in normalized coordinates3. Unnormalize:
Txx ~ xTx ~
THTH~1
ii Hxx
![Page 34: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/34.jpg)
Computing homography
• Assume we have matched points with outliers: How do we compute homography H?
Automatic Homography Estimation with RANSAC
![Page 35: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/35.jpg)
RANSAC: RANdom SAmple ConsensusScenario: We’ve got way more matched points than needed to fit the parameters, but we’re not sure which are correct
RANSAC Algorithm• Repeat N times
1. Randomly select a sample– Select just enough points to recover the parameters2. Fit the model with random sample
3. See how many other points agree• Best estimate is one with most agreement
– can use agreeing points to refine estimate
![Page 36: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/36.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 = 3 pixels
6. Repeat steps 2-5 N times– Choose H with most inliers
HZ Tutorial ‘99
ii Hxx
![Page 37: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/37.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. Project each image onto the same surface and blend
![Page 38: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/38.jpg)
Choosing a Projection Surface
Many to choose: planar, cylindrical, spherical, cubic, etc.
![Page 39: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/39.jpg)
Planar vs. Cylindrical Projection
Planar
Photos by Russ Hewett
![Page 40: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/40.jpg)
Planar vs. Cylindrical Projection
Planar
![Page 41: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/41.jpg)
Planar vs. Cylindrical Projection
Cylindrical
![Page 42: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/42.jpg)
Planar vs. Cylindrical Projection
Cylindrical
![Page 43: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/43.jpg)
Planar
Cylindrical
![Page 44: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/44.jpg)
Simple gain adjustment
![Page 45: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/45.jpg)
Automatically choosing images to stitch
![Page 46: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/46.jpg)
Recognizing Panoramas
Brown and Lowe 2003, 2007Some of following material from Brown and Lowe 2003 talk
![Page 47: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/47.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 48: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/48.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 49: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/49.jpg)
RANSAC for Homography
Initial Matched Points
![Page 50: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/50.jpg)
RANSAC for Homography
Final Matched Points
![Page 51: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/51.jpg)
Verification
![Page 52: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/52.jpg)
RANSAC for Homography
![Page 53: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/53.jpg)
Recognizing Panoramas (cont.)
(now we have matched pairs of images)4. Find connected components
![Page 54: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/54.jpg)
Finding the panoramas
![Page 55: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/55.jpg)
Finding the panoramas
![Page 56: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/56.jpg)
Finding the panoramas
![Page 57: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/57.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 58: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/58.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 59: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/59.jpg)
Bundle Adjustment
New images initialized with rotation, focal length of the best matching image
![Page 60: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/60.jpg)
Bundle Adjustment
New images initialized with rotation, focal length of the best matching image
![Page 61: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/61.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 62: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/62.jpg)
Multi-band Blending (Laplacian Pyramid)
• Burt & Adelson 1983– Blend frequency bands over range l
![Page 63: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/63.jpg)
Multiband blending
![Page 64: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/64.jpg)
Blending comparison (IJCV 2007)
![Page 65: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/65.jpg)
Blending Comparison
![Page 66: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/66.jpg)
Straightening
Rectify images so that “up” is vertical
![Page 67: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/67.jpg)
Further reading
Harley and Zisserman: Multi-view Geometry book• 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• Tutorial:
http://users.cecs.anu.edu.au/~hartley/Papers/CVPR99-tutorial/tut_4up.pdf
• Recognising Panoramas: Brown and Lowe, IJCV 2007 (also bundle adjustment)
![Page 68: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/68.jpg)
Things to remember
• Homography relates rotating cameras
• Recover homography using RANSAC and normalized DLT
• Can choose surface of projection: cylinder, plane, and sphere are most common
• Lots of room for tweaking (blending, straightening, etc.)
![Page 69: 10/19/10](https://reader035.vdocuments.us/reader035/viewer/2022062717/56812a7a550346895d8e051f/html5/thumbnails/69.jpg)
Next class
• Using interest points to find objects in datasets