1 interest operator lectures lecture topics –interest points 1 (linda) interest points,...
Post on 20-Dec-2015
225 views
TRANSCRIPT
![Page 1: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/1.jpg)
1
Interest Operator Lectures
• lecture topics
– Interest points 1 (Linda)• interest points, descriptors, Harris corners, correlation matching
– Interest points 2 (Linda)• Kadir operator, Fergus object recognition, Sivic video indexing
– Interest points 3 (Matt Brown from Microsoft Research: April 18)• improving feature matching and indexing, SIFT
– Image Stitching (Matt: April 23)• automatically combine multiple images to give seamless, “stitched” result
![Page 2: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/2.jpg)
2
Interest Operators
• Find “interesting” pieces of the image– e.g. corners, salient regions – Focus attention of algorithms– Speed up computation
• Many possible uses in matching/recognition– Search– Object recognition– Image alignment & stitching– Stereo– Tracking– …
![Page 3: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/3.jpg)
3
Interest points
0D structurenot useful for matching
1D structureedge, can be localised in 1D, subject to the aperture problem
2D structurecorner, or interest point, can be localised in 2D, good for matching
Interest Points have 2D structure. Edge Detectors e.g. Canny [Canny86] exist, but descriptors are more difficult.
![Page 4: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/4.jpg)
4
Local invariant photometric descriptors
( )local descriptor
Local : robust to occlusion/clutter + no segmentation
Photometric : (use pixel values) distinctive descriptions
Invariant : to image transformations + illumination changes
![Page 5: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/5.jpg)
5
History - Matching
1. Matching based on correlation alone 2. Matching based on geometric primitives
e.g. line segments
Not very discriminating (why?)
Solution : matching with interest points & correlation [ A robust technique for matching two uncalibrated images through
the recovery of the unknown epipolar geometry, Z. Zhang, R. Deriche, O. Faugeras and Q. Luong, Artificial Intelligence 1995 ]
![Page 6: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/6.jpg)
6
Approach
• Extraction of interest points with the Harris detector
• Comparison of points with cross-correlation
• Verification with the fundamental matrix
![Page 7: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/7.jpg)
7
Harris detector
Interest points extracted with Harris (~ 500 points)
![Page 8: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/8.jpg)
8
Harris detector
![Page 9: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/9.jpg)
9
Harris detector
![Page 10: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/10.jpg)
10
Cross-correlation matching
Initial matches – motion vectors (188 pairs)
![Page 11: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/11.jpg)
11
Global constraints
Robust estimation of the fundamental matrix (RANSAC)
99 inliers 89 outliers
![Page 12: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/12.jpg)
12
Summary of the approach
• Very good results in the presence of occlusion and clutter– local information– discriminant greyvalue information – robust estimation of the global relation between images– works well for limited view point changes
• Solution for more general view point changes– wide baseline matching (different viewpoint, scale and rotation)– local invariant descriptors based on greyvalue information
![Page 13: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/13.jpg)
13
Invariant Features
• Schmid & Mohr 1997, Lowe 1999, Baumberg 2000, Tuytelaars & Van Gool 2000, Mikolajczyk & Schmid 2001, Brown & Lowe 2002, Matas et. al. 2002, Schaffalitzky & Zisserman 2002
![Page 14: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/14.jpg)
14
Approach
1) Extraction of interest points (characteristic locations)
2) Computation of local descriptors (rotational invariants)
3) Determining correspondences
4) Selection of similar images
( )local descriptor
interest points
![Page 15: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/15.jpg)
15
Harris detector
Based on the idea of auto-correlation
Important difference in all directions => interest point
![Page 16: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/16.jpg)
16
Autocorrelationa
Autocorrelation function (ACF) measures
the self similarity of a signal
![Page 17: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/17.jpg)
17
Autocorrelation
a
a
a
• Autocorrelation related to sum-square difference:
(if ∫f(x)2dx = 1)
![Page 18: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/18.jpg)
18
Background: Moravec Corner Detector
w
• take a window w in the image• shift it in four directions
(1,0), (0,1), (1,1), (-1,1)• compute a difference for each• compute the min difference at
each pixel• local maxima in the min image
are the corners
E(x,y) = ∑ w(u,v) |I(x+u,y+v) – I(u,v)|2 u,v in w
![Page 19: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/19.jpg)
19
Shortcomings of Moravec Operator
• Only tries 4 shifts. We’d like to consider “all” shifts.
• Uses a discrete rectangular window. We’d like to use a smooth circular (or later elliptical) window.
• Uses a simple min function. We’d like to characterize variation with respect to direction.
Result: Harris Operator
![Page 20: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/20.jpg)
20
Harris detector
2
),(
)),(),((),( yyxxIyxIyxf kkkWyx
k
kk
y
xyxIyxIyxIyyxxI kkykkxkkkk )),(),((),(),(with
2
),(
),(),(),(
Wyx
kkykkx
kky
xyxIyxIyxf
Auto-correlation fn (SSD) for a point and a shift ),( yx ),( yx
Discrete shifts can be avoided with the auto-correlation matrix
what is this?
![Page 21: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/21.jpg)
21
Harris detector
Rewrite as inner (dot) product
The center portion is a 2x2 matrix
![Page 22: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/22.jpg)
22
Harris detector
y
x
yxIyxIyxI
yxIyxIyxI
yx
Wyxkky
Wyxkkykkx
Wyxkkykkx
Wyxkkx
kkkk
kkkk
),(
2
),(
),(),(
2
)),((),(),(
),(),()),((
Auto-correlation matrix M
![Page 23: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/23.jpg)
23
Harris detection
• Auto-correlation matrix– captures the structure of the local neighborhood– measure based on eigenvalues of M
• 2 strong eigenvalues => interest point
• 1 strong eigenvalue => contour
• 0 eigenvalue => uniform region
• Interest point detection– threshold on the eigenvalues– local maximum for localization
![Page 24: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/24.jpg)
24
Some Details from the Harris Paper
• Corner strength R = Det(M) – k Tr(M)2
• Let and be the two eigenvalues• Tr(M) = + • Det(M) =
• R is positive for corners, - for edges, and small for flat regions
• Select corner pixels that are 8-way local maxima
![Page 25: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/25.jpg)
25
Determining correspondences
Vector comparison using a distance measure
( ) ( )=?
What are some suitable distance measures?
![Page 26: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/26.jpg)
26
Distance Measures
• We can use the sum-square difference of the values of the pixels in a square neighborhood about the points being compared.
SSD = ∑∑(W1(i,j) –(W2(i,j))2
W1 W2
![Page 27: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/27.jpg)
27
Some Matching Results
![Page 28: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/28.jpg)
28
Some Matching Results
![Page 29: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/29.jpg)
29
Summary of the approach
• Basic feature matching = Harris Corners & Correlation• Very good results in the presence of occlusion and clutter
– local information– discriminant greyvalue information – invariance to image rotation and illumination
• Not invariance to scale and affine changes
• Solution for more general view point changes– local invariant descriptors to scale and rotation– extraction of invariant points and regions
![Page 30: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/30.jpg)
30
Rotation/Scale Invariance
Translation Rotation Scale
Is Harris invariant?
? ? ?
Is correlation invariant?
? ? ?
original translated rotated scaled
![Page 31: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/31.jpg)
31
Rotation/Scale Invariance
Translation Rotation Scale
Is Harris invariant?
? ? ?
Is correlation invariant?
? ? ?
original translated rotated scaled
![Page 32: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/32.jpg)
32
Rotation/Scale Invariance
Translation Rotation Scale
Is Harris invariant?
YES ? ?
Is correlation invariant?
? ? ?
original translated rotated scaled
![Page 33: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/33.jpg)
33
Rotation/Scale Invariance
Translation Rotation Scale
Is Harris invariant?
YES YES ?
Is correlation invariant?
? ? ?
original translated rotated scaled
![Page 34: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/34.jpg)
34
Rotation/Scale Invariance
Translation Rotation Scale
Is Harris invariant?
YES YES NO
Is correlation invariant?
? ? ?
original translated rotated scaled
![Page 35: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/35.jpg)
35
Rotation/Scale Invariance
Translation Rotation Scale
Is Harris invariant?
YES YES NO
Is correlation invariant?
? ? ?
original translated rotated scaled
![Page 36: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/36.jpg)
36
Rotation/Scale Invariance
Translation Rotation Scale
Is Harris invariant?
YES YES NO
Is correlation invariant?
YES ? ?
original translated rotated scaled
![Page 37: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/37.jpg)
37
Rotation/Scale Invariance
Translation Rotation Scale
Is Harris invariant?
YES YES NO
Is correlation invariant?
YES NO ?
original translated rotated scaled
![Page 38: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/38.jpg)
38
Rotation/Scale Invariance
Translation Rotation Scale
Is Harris invariant?
YES YES NO
Is correlation invariant?
YES NO NO
original translated rotated scaled
![Page 39: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/39.jpg)
39
Invariant Features
• Local image descriptors that are invariant (unchanged) under image transformations
![Page 40: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/40.jpg)
40
Canonical Frames
![Page 41: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/41.jpg)
41
Canonical Frames
![Page 42: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/42.jpg)
42
Multi-Scale Oriented Patches
• Extract oriented patches at multiple scales
![Page 43: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/43.jpg)
43
Multi-Scale Oriented Patches
• Sample scaled, oriented patch
![Page 44: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/44.jpg)
44
Multi-Scale Oriented Patches
• Sample scaled, oriented patch– 8x8 patch, sampled at 5 x scale
![Page 45: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/45.jpg)
45
• Sample scaled, oriented patch– 8x8 patch, sampled at 5 x scale
• Bias/gain normalised– I’ = (I – )/
Multi-Scale Oriented Patches
8 pixels
40 pixels
![Page 46: 1 Interest Operator Lectures lecture topics –Interest points 1 (Linda) interest points, descriptors, Harris corners, correlation matching –Interest points](https://reader035.vdocuments.site/reader035/viewer/2022062407/56649d4a5503460f94a264c8/html5/thumbnails/46.jpg)
46
Matching Interest Points: Summary
• Harris corners / correlation– Extract and match repeatable image features– Robust to clutter and occlusion– BUT not invariant to scale and rotation
• Multi-Scale Oriented Patches– Corners detected at multiple scales– Descriptors oriented using local gradient
• Also, sample a blurred image patch
– Invariant to scale and rotation
Leads to: SIFT – state of the art image features