Title: Local invariant features
1Local invariant features
- Cordelia Schmid
- INRIA, Grenoble
2Overview
- Introduction to local features
- Harris interest points SSD, ZNCC, SIFT
- Scale affine invariant interest point detectors
- Evaluation and comparison of different detectors
- Region descriptors and their performance
3Introduction
- Local invariant photometric features
local descriptor
Local robust to occlusion/clutter no object
segmentation Photometric distinctive Invariant
to image transformations illumination changes
4Partial visibility/occlusion
5Clutter (additional objects)
6Image transformation rotation
7Image transformations scale change
8Illumination variations
9Viewpoint changes
10Local features - history
- Line segments Lowe87, Ayache90
- Interest points cross correlation Z. Zhang et
al. 95 - Rotation invariance with differential invariants
SchmidMohr96 - Scale affine invariant detectors Lindeberg98,
Lowe99, TuytelaarsVanGool00,
MikolajczykSchmid02, Matas et al.02 - Dense detectors and descriptors LeungMalik99,
Fei-Fei Perona05, Lazebnik et al.06 - Contour and region (segmentation) descriptors
Shotton et al.05, Opelt et al.06, Ferrari et
al.06, Leordeanu et al.07
11Local features
- 1) Extraction of local features
- Contours/segments
- Interest points regions
- Regions by segmentation
- Dense features, points on a regular grid
- 2) Description of local features
- Dependant on the feature type
- Segments ? angles, length ratios
- Interest points ? greylevels, gradient histograms
- Regions (segmentation) ? texture color
distributions
12Local features Contours/segments
13Local features interest points
14Local features segmentation
15Application Matching
Find corresponding locations in the image
16Application Image retrieval
Search for images with the same/similar object in
a set of images
17Overview
- Introduction to local features
- Harris interest points SSD, ZNCC, SIFT
- Scale affine invariant interest point detectors
- Evaluation and comparison of different detectors
- Region descriptors and their performance
18Harris detector Harris Stephens88
Based on the idea of auto-correlation
Important difference in all directions gt
interest point
19Harris detector
Auto-correlation function for a point
and a shift
20Harris detector
Auto-correlation function for a point
and a shift
small in all directions
? uniform region
? contour
large in one directions
? interest point
large in all directions
21 Harris detector
22Harris detector
Discret shifts are avoided based on the
auto-correlation matrix
23Harris detector
Auto-correlation matrix
24Harris detector
- Auto-correlation matrix
- captures the structure of the local neighborhood
- measure based on eigenvalues of this matrix
- 2 strong eigenvalues
- 1 strong eigenvalue
- 0 eigenvalue
gt interest point
gt contour
gt uniform region
25Harris eigenvalues
26Harris detector
- Interest point detection
- Treshold (absolut, relatif, number of corners)
- Local maxima
27Harris - invariance to transformations
- Geometric transformations
- translation
- rotation
- similitude (rotation scale change)
- affine (valide for local planar objects)
- Photometric transformations
- Affine intensity changes (I ? a I b)
28Harris detector
Interest points extracted with Harris ( 500
points)
29Comparison of patches - SSD
Comparison of the intensities in the neighborhood
of two interest points
image 2
image 1
SSD sum of square difference
Small difference values signifies similar patches
30Comparison of patches
SSD
Invariance to photometric transformations?
Intensity changes (I ? I b)
Intensity changes (I ? aI b)
31Cross-correlation ZNCC
zero normalized SSD
ZNCC zero normalized cross correlation
ZNCC values between -1 and 1, 1 when identical
patches in practice threshold around 0.5
32Cross-correlation matching
Initial matches (188 pairs)
33Global constraints
Robust estimation of the fundamental matrix
99 inliers
89 outliers
34Local descriptors
- Greyvalue derivatives
- Differential invariants Koen87
- SIFT descriptor Lowe99
- Moment invariants Van Gool et al.96
- Shape context Belongie et al.02
35Greyvalue derivatives Image gradient
- The gradient points in the direction of most
rapid increase in intensity
- The gradient direction is given by
- how does this relate to the direction of the
edge? - The edge strength is given by the gradient
magnitude
Source Steve Seitz
36Differentiation and convolution
- Recall, for 2D function, f(x,y)
- We could approximate this as
- Convolution with the filter
Source D. Forsyth, D. Lowe
37Finite difference filters
- Other approximations of derivative filters exist
Source K. Grauman
38Effects of noise
- Consider a single row or column of the image
- Plotting intensity as a function of position
gives a signal
Source S. Seitz
39Solution smooth first
f
Source S. Seitz
40Derivative theorem of convolution
- Differentiation is convolution, and convolution
is associative - This saves us one operation
Source S. Seitz
41Local descriptors
- Greyvalue derivatives
- Simple difference filters (-1,1)
- Convolution with Gaussian derivatives
42Local descriptors rotation invariance
Notation for greyvalue derivatives
43Local descriptors rotation invariance
- Invariance to image rotation differential
invariants Koen87
gradient magnitude
Laplacian
44Laplacian of Gaussian (LOG)
45Local descriptors - rotation invariance
- Estimation of the dominant orientation
- extract gradient orientation
- histogram over gradient orientation
- peak in this histogram
- Rotate patch in dominant direction
46Local descriptors illumination change
- Robustness to illumination changes
in case of an affine transformation
- Normalization of derivatives with gradient
magnitude
- Normalization of the image patch with mean and
variance
47SIFT descriptor Lowe99
- Approach
- 8 orientations of the gradient
- 4x4 spatial grid
- soft-assignment to spatial bins, dimension 128
- normalization of the descriptor to norm one
- comparison with Euclidean distance
3D histogram
image patch
gradient
x
?
?
y
48Invariance to scale changes
- Scale change between two images
- Scale factor s can be eliminated
- Support region for calculation!!
- In case of a convolution with Gaussian
derivatives defined by