segmentation and enhancement of latent …cse902/s14/ppt/latent segmentation and...segmentation and...
TRANSCRIPT
![Page 1: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/1.jpg)
Segmentation and Enhancement of Latent
Fingerprints: A Coarse to Fine Ridge
Structure Dictionary
Kai Cao
January 16, 2014
![Page 2: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/2.jpg)
Fingerprint
D. Maltoni et al., Handbook of Fingerprint Recognition, Springer, 2003.
Fingerprint Image
![Page 3: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/3.jpg)
Fingerprint level-1 features
1) Pattern type
2) Orientation field
3) Singular points
4) Frequency field
Left loop Right loop Whorl
Arch Tented arch
Core
Delta
![Page 4: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/4.jpg)
Fingerprint level-2 features
• Minutiae: ridge endings and bifurcations
Ridge ending Ridge bifurcation
![Page 5: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/5.jpg)
Incipient ridge Pores
• Include all dimensional attributes of the ridge such as pores, edge contour, incipient ridges and other permanent details
Edge contour
Fingerprint level-3 features
![Page 6: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/6.jpg)
Fingerprint types
Rolled fingerprint Plain fingerprint latent fingerprint
Anil K. Jain and Jianjiang Feng, Latent fingerprint matching, TPAMI, 2011
![Page 7: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/7.jpg)
Challenges in latent matching
Unclear ridges Partial fingerprint
Large distortion Complex background
Reliable
feature
extraction
Robust
feature
matching
![Page 8: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/8.jpg)
Essential modules for feature
extraction
Latent fingerprint Segmented
fingerprint
Frequency field
Orientation field
Segmented and
enhanced fingerprint
Gabor filter
![Page 9: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/9.jpg)
Algorithm flowchart
Offline stage
Offline stage
Online stage
![Page 10: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/10.jpg)
Fingerprints for
training
32×32 patches
for training
64×64 Patches
for training Dictionary
learning
Dictionary
learning
1 coarse-level
dictionary
16 fine-level
dictionaries
Provides coarse-
level quality map
with orientation field
and frequency field
Provides fine-level quality
map with orientation field
and frequency field
Friction ridge dictionary learning
10
(0, π/16]
(π/16, 2π/16]
(15π/16, π]
(1) Fingerprint selection: NFIQ index <= 2
(2) Patch selection: average quality >=3.75
![Page 11: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/11.jpg)
Friction ridge dictionary learning
Initialize Dictionary D
Sparse Coding
Orthogonal Matching Pursuit
Update Dictionary
One atom at a time
• Learning algorithm: K-SVD[1]
• Database: High quality patches in NIST DB 4
[1] M. Aharon, M. Elad and A. Bruchstein, “K-SVD: An Algorithm for Designing Overcomplete
Dictionaries for Sparse Representation”, IEEE TSP, 54(11): 4311-4322, 2006. 11
Objective function:
Each example has a
sparse
representation
![Page 12: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/12.jpg)
Latent fingerprint decomposition
• local total variation[1] (LTV)
• relative reduction rate of LTV
[1] A. Buades, T. Le, J.-M. Morel, and L. Vese, “Fast cartoon + texture image filters,” IEEE
Transactions on Image Processing, 19(8):1978 –1986, 2010.
( ) *| | ( ),LTV f L f x
Background
patch
Fingerprint
patch
![Page 13: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/13.jpg)
= +
Texture part Cartoon part Gray image
• Cartoon part and texture part
Cartoon part
Texture part
Latent fingerprint decomposition
![Page 14: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/14.jpg)
Patch reconstruction • Patch reconstructed with T dictionary elements
• Structure similarity (SSIM[2]) measures the quality
(a) (b) (c) (d) (e)
SSIM=0.52 SSIM=0.59 SSIM=0.63 SSIM=0.65
SSIM=0.17 SSIM=0.28 SSIM=0.33 SSIM=0.37
SSIM=0.01 SSIM=0.01 SSIM=0.02 SSIM=0.02
Quality estimation
[2] Wang et al., “Image Quality Assessment: From Error Visibility to Structural Similarity”, IEEE
Transaction on Image Processing, 13(4): 600-612, 2004.
1 2
2 2 2 2
1 2
2( , )
x y x y
x y x y
C CSSIM x y
C C
T=1 T=2 T=3 T=4
![Page 15: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/15.jpg)
Patch reconstruction • Patch reconstructed by sparse coding
• Structure similarity (SSIM[2]) measures the quality
• Orientation and frequency fields are estimated
from the reconstructed patch
Orientation field estimation
[2] Wang et al., “Image Quality Assessment: From Error Visibility to Structural Similarity”, IEEE
Transaction on Image Processing, 13(4): 600-612, 2004.
T=1 T=2 T=3 T=4
![Page 16: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/16.jpg)
Sparsity control
SSIM=0.52
SSIM=0.18
SSIM=0.01
Patch Most similar
dictionary
element
Orientation
field
Advantages of T=1:
1) A single dictionary element is good enough
to recover level-1 features (orientation and
frequency)
2) No matrix inversion needed in orthogonal
matching pursuit
3) Orientation and frequency fields of the
dictionary elements can be estimated
offline
![Page 17: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/17.jpg)
Coarse to fine strategy
Latent Coarse-level
quality
Fine-level quality
• Why do we need a coarse to fine strategy?
Patch
32*32
Most similar
dictionary
element Patch
64*64
Most similar
dictionary
element
1) Small patch is easy affected by background noise
2) Large patch is not able to capture details
![Page 18: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/18.jpg)
Coarse to fine strategy • How does coarse to fine strategy work?
Patch
64*64
Most similar
dictionary
element
(0, π/16]
(π/16, 2π/16]
(15π/16, π]
Patch
32*32
Most similar
dictionary
element from
selected
dictionary
![Page 19: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/19.jpg)
Some results on NIST SD27
Good
latent
Bad
latent
Ugly
latent
(a) Gray image (b) Texture image (c) Coarse quality map (d) Fine quality map
![Page 20: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/20.jpg)
Some results on NIST SD27
Good
latent
Bad
latent
Ugly
latent
(a) Gray image (b) Texture image (c) Segmentation result (d) Segmentation and
enhancement result
![Page 21: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/21.jpg)
Performance evaluation
21
• Latent Database: 258 latents in NIST SD27 and 449 latents in WVU DB
• Background Database : 27,258 rolled prints in NIST SD27 (258) and NIST
SD14 (27,000)
• Matcher: Commercial off the shelf (COTS) tenprint matcher
[3]J. Feng, J. Zhou, and A. K. Jain, “Orientation field estimation for latent fingerprint enhancement,” IEEE TPAMI,
vol. 54, no. 4, pp. 925–940, 2013.
[4] J. Zhang, R. Lai, and C.-C. Kuo, “Adaptive directional total variation model for latent fingerprint segmentation,”
IEEE TIFS, vol. 8, no. 8, pp. 1261–1273, 2013.
Four latent image types (templates) that can be input to a COTS matcher. (a) Original latent image (G084 from
NIST SD 27), (b) latent segmented and enhanced by the proposed algorithm, (c) latent segmented by the
proposed segmentation algorithm, but enhanced by the algorithm in [3] and (d) latent segmented and enhanced
by the ADTV model [4], (d) the true mate (rolled print). The retrieved ranks of the mated rolled print by COTS for
(a), (b), (c) and (d) are 5536, 1, 66 and 12462, respectively.
![Page 22: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/22.jpg)
Performance evaluation
22
• Latent Database: 258 latents in NIST SD27 and 449 latents in WVU DB
• Background Database : 27,258 rolled prints in NIST SD27 (258) and NIST
SD14 (27,000)
• Matcher: Commercial off the shelf (COTS) tenprint matcher
[3]J. Feng, J. Zhou, and A. K. Jain, “Orientation field estimation for latent fingerprint enhancement,” IEEE TPAMI,
vol. 54, no. 4, pp. 925–940, 2013.
[4] J. Zhang, R. Lai, and C.-C. Kuo, “Adaptive directional total variation model for latent fingerprint segmentation,”
IEEE TIFS, vol. 8, no. 8, pp. 1261–1273, 2013.
(a) NIST SD27 (b) WVU DB
![Page 23: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/23.jpg)
Performance evaluation • Latent Database: 258 latents in NIST SD27 and 449 latents in WVU DB
• Background Database : 27,258 rolled prints in NIST SD27 (258) and NIST
SD14 (27,000)
• Matcher: Commercial off the shelf (COTS) latent matcher
(a) NIST SD27 (b) WVU DB
![Page 24: Segmentation and Enhancement of Latent …cse902/S14/ppt/Latent segmentation and...Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary](https://reader034.vdocuments.us/reader034/viewer/2022042022/5e7941a5ad287967d704a36b/html5/thumbnails/24.jpg)
Thanks!