Image Matching using
Local Symmetry
Features
Daniel Cabrini Hauagge
Noah Snavely
Cornell University
1
Outline






Introduction
Local Symmetry
Scale Space
Local Symmetry Features
Experimental Results
Conclusion
2
Introduction

Analysis of symmetry :
A long –standing problem in computer vision.
3
Introduction
4
Local Symmetry

Method- takes an image as input, and
computes local symmetry scores over the
image and across scale space.


Scoring local symmetries
Gradient histogram-based score
5
Local Symmetry

Bilateral symmetry

2n-fold rotational symmetry

For both symmetry types
6
Local Symmetry
7
Scoring local symmetries

Define our score

Symmetry type


Distance function


If the image f exhibits perfect symmetry types at
location p, then f(q) = f(Ms,p(q)) for all q.
d(q,r)= | f(q)-f(r) |
Weight mask

gives the importance of each set of corresponding
point pairs around the center point p in determining
the symmetry score at p.
8
Scoring local symmetries

Define a function that we call the local
symmetry distance

SS = L*SD
9
Gradient histogram-based
score
10
Scale space

For the purpose of feature detection, we choose a
different function for

r is the distance to the point of interest
A is a normalization constant
r0 is the ring radius
ψ controls the width of the ring



11
Scale space
12
Local Symmetry Features

By finding local maxima of the score, and
feature description, by building a feature
vector from local symmetry scores


Feature detector
Feature descriptor
13
Feature detector


Using the SYM-IR function as a detector.
Represent the support of each feature as a
circle of radius s centered at (x,y) in the
original image.
14
Feature detector
15
Feature detector
16
Feature descriptor

SYMD encodes the distribution of the three
SYM-I scores around a feature location at the
detected scale
17
Experimental Results

Evaluate the detector, comparing its
repeatability to that of the common DoG and
MSER detectors.


Evaluating detections
Evaluating descriptors
18
Evaluating detections


For each image pair(I1,I2) in the dataset, and
each detector, we detect sets of keypoints K1
and K2 , and compare these detections using
the known homography H12 mapping points
from I1 to I2.
K1 is rescaled by a factor s so that it has a
fixed area A ; we denote this scaled detection
K1S also applied to K2. Finally, the relative
overlap of the support regions of H12K1S and
K2 gives an overlap score.
19
Evaluating detections
20
Evaluating detections
21
Evaluating descriptors

A precision-call curve that summarizes the
quality of the match scores.
22
Evaluating descriptors

We generate two sets of perfectly matched
synthetic detections by creating a set of
keypoints K1 on a grid in I1 (in our
experiments the spacing between points is 25
pixels and the scale of each keypoint is set to
6.25). We then map these keypoints to I2
using H12, creating a matched set of keys K2.
We discard keypoints whose support regions
are not fully within the image.
23
Evaluating descriptors
24
Evaluating descriptors
25
Conclusion

To evaluate our method, we created a new
dataset of image pairs with dramatic
appearance changes, and showed that our
features are more repeatable, and yield
complementary information, compared to
standard features such as SIFT.
26

Thanks for your listening.
27