complex zernike moments features for shape-based image retrieval
DESCRIPTION
Complex Zernike Moments Features for Shape-Based Image Retrieval. IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 39, NO. 1, JANUARY 2009. 指導教授:李育強 報告者 :楊智雁 日期 : 2010/03/15. Outline. 1. Introduction. Zernike Moments. 2. Zm Phase and Magnitude. 3. - PowerPoint PPT PresentationTRANSCRIPT
南台科技大學 資訊工程系
Complex Zernike Moments Features for
Shape-Based Image Retrieval
指導教授:李育強報告者 :楊智雁日期 : 2010/03/15
IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 39, NO. 1, JANUARY 2009
2
Outline
Introduction1
Zernike Moments2
Zm Phase and Magnitude3
Experiments4
5 Conclusion
3
1. Introduction
Existing CBIR systems can be broadly categorized into two groups
Contour and region-based descriptors
As the most commonly used approaches for region-based shape descriptors (geometric moments)
4
1. Introduction (c.)
However, geometric moments do not have any of the desired invariance
Such as translation, scale, or rotation invariance
In this paper, we try to find out the relative importance of the phase and magnitude of Zernike Moments
5
2. Zernike Moments
2/)(
0
2),,()(mn
s
snnm smncR
Basis Function
Radial polynomials
)!2/(()!2/)((!
)!()1(),,(
smnsmns
snsmnc s
)exp()(),( jmRV nmnm
6
2. Zernike Moments (c.)
Zernike moments measurement
2
0
1
0
* ),(),(1
ddVfn
Z nmnm
7
3. Zm Phase and Magnitude
The reconstructed images have far less resemblance to the original image than those by using both magnitude and phase components
2
12 )))(((
))(exp()(),(
n mnmnm
n mnmnmnm
Rc
mjRcI
8
3. Zm Phase and Magnitude (c.)
),(),(),(),( '1,0
' nmRn
Rnm
Rnm m
The corrected phase angle of the rotated image is the same as the corrected phase angle ofthe nonrotated image
),(' Rnm
),(' nm
9
3. Zm Phase and Magnitude (c.)
Angle-based distance and magnitude-based distance
N
iangang id
ND
1
2 )(1
N
imagmag id
ND
1
2 )(1
magmagangang DDD
10
4. Experiments
A. Preparation for Test DBs Scale test DB 、 Rotation test DB 、 Subject test DB 、
Noisy test DB
B. Measurement of Retrieval Performance 1. P−R Graph 2. BEP
11
4. Experiments (c.)
C. Experiment Results 1. IZMD With Different Max Orders
12
4. Experiments (c.)
2. Performance Comparison of IZMD and ZMD
13
4. Experiments (c.)
14
4. Experiments (c.)
3. Performance Comparison of IZMD and GFD
15
4. Experiments (c.)
16
5.Conclusion
Rotation 、 translation 、 scaling or change in viewpoint , We propose here IZMD for robust image retrieval
Its superior performance in noise robustness and subject discriminability when compared with magnitude-only ZMD
We would incorporate object segmentation techniques into the proposed IZMD framework
南台科技大學 資訊工程系