Dermoscopic Interest Point Detector and Descriptor Howard Zhou1, Mei Chen2, James M.
Download
Report
Transcript Dermoscopic Interest Point Detector and Descriptor Howard Zhou1, Mei Chen2, James M.
Dermoscopic Interest Point
Detector and Descriptor
Howard Zhou1, Mei Chen2, James M. Rehg1
1School
of Interactive Computing, Georgia Tech
2Intel Research Pittsburgh
Skin cancer
Skin cancer : most common type of cancer ( > 1 million )
Skin
Lung
Breast
Prostate
Bladder
0
200,000
400,000
600,000
800,000
1,000,000 1,200,000
[ Top 5 categories of estimated annual cancer incidence for 2009 from National Cancer Institute ]
2009-07-01
2
Skin cancer
Skin cancer : most common type of cancer ( > 1 million )
forms in tissues of the skin
Skin lesions
2009-07-01
3
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Skin cancer
Skin cancer : most common type of cancer ( > 1 million )
forms in tissues of the skin
Benign lesions
2009-07-01
Skin cancer
4
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Skin cancer
Skin cancer : most common type of cancer ( > 1 million )
forms in tissues of the skin
Benign lesions
Skin cancer
Basal cell
carcinoma
Squamous cell
carcinoma
Melanoma
2009-07-01
5
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopy
Non-invasive imaging technique
Improve diagnostic accuracy by 30%
Skin cancer
Basal cell
carcinoma
Squamous cell
carcinoma
Melanoma
2009-07-01
6
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopy
Non-invasive imaging technique
Improve diagnostic accuracy by 30%
Clinical view
2009-07-01
7
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopy
Non-invasive imaging technique
Improve diagnostic accuracy by 30%
Microscope + light + liquid medium
Dermatoscope
2009-07-01
8
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopy
Non-invasive imaging technique
Improve diagnostic accuracy by 30%
Microscope + light + liquid medium
Reveal pigmented structures
Dermatoscope
Dermoscopy view
2009-07-01
9
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopic features
Pigmented structures revealed by dermoscopy
Dermoscopy view
2009-07-01
10
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopic features
Pigmented structures revealed by dermoscopy
Dermoscopy view
Blue-white veil
2009-07-01
11
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopic features
Pigmented structures revealed by dermoscopy
Dermoscopy view
Blue-white veil
Scar-like
depigmentation
2009-07-01
12
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopic features
Pigmented structures revealed by dermoscopy
Dermoscopy view
Blue-white veil
Scar-like
depigmentation
Brown globules
2009-07-01
13
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopic features
Pigmented structures revealed by dermoscopy
Dermoscopy view
Blue-white veil
Scar-like
depigmentation
Brown globules
Negative network
2009-07-01
14
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopic features
Pigmented structures revealed by dermoscopy
[Betta et al. 2006], [Grana et al. 2006], [Iyatomi et al. 2007],…
Dermoscopy view
Blue-white veil
Scar-like
depigmentation
Brown globules
Negative network
2009-07-01
15
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopic features
Over 100 dermoscopic features
Dermoscopy view
Blue-white veil
Scar-like
depigmentation
Brown globules
Negative network
…
2009-07-01
16
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopic features
Over 100 dermoscopic features
Multiple binary classifiers for each image
Dermoscopy view
Blue-white veil
BW classifier
Scar-like
depigmentation
SLD classifier
Brown globules
BG classifier
Negative network NN classifier
…
…
2009-07-01
17
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopic features
General detector?
Dermoscopy view
Blue-white veil
Scar-like
depigmentation
Brown globules
Negative network
…
2009-07-01
18
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopic features
General detector?
Dermoscopy view
Blue-white veil
Scar-like
depigmentation
Brown globules
Negative network
…
Dermoscopic features consist of low level image characteristics
(ridges, blobs, streaks, pigmentation,…)
2009-07-01
19
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopic features
General detector?
Dermoscopy view
Blue-white veil
Scar-like
depigmentation
Brown globules
Negative network
…
Dermoscopic features consist of low level image characteristics
(ridges, blobs, streaks, pigmentation,…)
interest points
2009-07-01
20
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopic Interest Point (DIP)
General detector: concentration/configuration of interest points
bag-of-visual-words approach
Dermoscopy view
Blue-white veil
Scar-like
depigmentation
Brown globules
Negative network
…
Dermoscopic features consist of low level image characteristics
(ridges, blobs, streaks, pigmentation,…)
interest points
2009-07-01
21
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ]
Dermoscopic Interest Point (DIP)
Inspired by general interest point detector and descriptors
We propose Dermoscopic Interest Point (DIP)
(SIFT & SURF)
detector - to extract these low level building blocks
descriptor – for constructing a general visual vocabulary for
dermoscopic features
2009-07-01
22
Dermoscopic Interest Point (DIP)
Compared to the general interest point detector and
descriptors (SIFT & SURF)
Same key issues
Similar to SIFT & SURF
Repeatable
Distinctive
Robust to noise and deformation (geometric and photometric)
Corners and blobs
Scale and rotation invariant
In addition
Curvilinear features (fibrillar pattern and radial streaming)
Color component
2009-07-01
23
Detector
Corners and blobs
Fast-Hessian detector [Bay, et al. 2006]
Hessian matrix
2009-07-01
24
Detector
Corners and blobs
Fast-Hessian detector [Bay, et al. 2006]
Hessian matrix
Box filter approximation to replace Gaussian derivatives
Fast using Integral image
2009-07-01
25
Detector
Corners and blobs
Fast-Hessian detector [Bay, et al. 2006]
Hessian matrix
Curvilinear structures
Curvilinear detector [Steger, 1996]
2009-07-01
26
Detector
Corners and blobs
Fast-Hessian detector [Bay, et al. 2006]
Hessian matrix
Curvilinear structures
Curvilinear detector [Steger, 1996]
2009-07-01
27
Detector
Corners and blobs
Fast-Hessian detector [Bay, et al. 2006]
Hessian matrix
Curvilinear structures
Curvilinear detector [Steger, 1996]
2009-07-01
28
Detector
Corners and blobs
Fast-Hessian detector [Bay, et al. 2006]
Hessian matrix
Curvilinear structures
Curvilinear detector [Steger, 1996]
2009-07-01
29
Descriptor
Distinctiveness
Spatially localized information
Distribution of gradient-related features
Dermscopic: color features
Invariance (Repeatability)
Relative strength to reduce the effect of photometric changes
Relative orientation for rotation invariance
2009-07-01
30
Descriptor
Distinctiveness
Invariance (Repeatability)
Spatially localized information
Distribution of gradient-related features
Dermscopic: color features
Relative strength to reduce the effect of photometric changes
Relative orientation for rotation invariance
To construct
Reproducible orientation
2009-07-01
31
Descriptor
Distinctiveness
Invariance (Repeatability)
Spatially localized information
Distribution of gradient-related features
Dermscopic: color features
Relative strength to reduce the effect of photometric changes
Relative orientation for rotation invariance
To construct
Reproducible orientation
2009-07-01
32
Descriptor
Distinctiveness
Invariance (Repeatability)
Spatially localized information
Distribution of gradient-related features
Dermscopic: color features
Relative strength to reduce the effect of photometric changes
Relative orientation for rotation invariance
To construct
Reproducible orientation
Feature vector
2009-07-01
33
Descriptor
Orientation
For rotation invariance
Haar-wavelet responses in x and y
direction (in a circular neighborhood)
2009-07-01
34
Descriptor
Orientation
For rotation invariance
Haar-wavelet responses in x and y
direction (in a circular neighborhood)
Reponses represented as 2D vectors
dy
dx
2009-07-01
35
Descriptor
Orientation
For rotation invariance
Haar-wavelet responses in x and y
direction (in a circular neighborhood)
Reponses represented as 2D vectors
dy
Average responses in a sliding window
of 60 degree
dx
2009-07-01
36
Descriptor
Orientation
For rotation invariance
Haar-wavelet responses in x and y
direction (in a circular neighborhood)
Reponses represented as 2D vectors
dy
Average responses in a sliding window
of 60 degree
dx
The longest vector indicates the
orientation
2009-07-01
37
Descriptor
Descriptor components
Context of the descriptor: a square region oriented along the
orientation (centered around the interest point)
Local statistics
2009-07-01
Uniform 4 x 4 subregions
Intensity gradients (I): Sum of Haar-wavelet responses: dx, dy, |dx|,
|dy|
Color statistics (C): Coarse color histogram of the region (alpha &
beta channels in L*a*b space)
[ Image courtesy of Bay et al. 2006]
38
Dermoscopy Interest Point
2009-07-01
39
Dermoscopy specific
Common interest point descriptor ignores linear features
SURF
2009-07-01
DIP
40
Experiment
2009-07-01
41
Conclusion
A generalized framework for characterizing dermoscopic
features using Dermoscopic Interest Point (DIP)
A feature detector and a descriptor specifically designed for
this purpose
Initial experiments showed that our scheme achieves a
comparable level of invariance to lighting, scale, and rotation
changes
2009-07-01
42
Future work
Build a vocabulary of dermoscopic features using DIP
Explore the possibility of using DIP in skin CAD related
applications:
Dermoscopic feature extraction and classification
Dermoscopy image registration
Dermoscopy image search and retrieval via dermoscopic
features
2009-07-01
43
Acknowledgement
Collaborators
(in alphabetical order)
Dr. Laura K. Ferris M.D. Ph.D. UPMC
Richard Gass, Intel Research Pittsburgh
Casey Helfrich, Intel Research Pittsburgh
Many thanks to our anonymous reviewers for their
helpful comments and suggestion
2009-07-01
44
Thank you
Thank you
!
2009-07-01
45
Related publications
Interest pointer detector and descriptors
Distinctive image features from scale-invariant
keypoints
David G. Lowe
Intl. J. of Computer Vision (IJCV), 2004
Surf: Speeded up robust features
Herbert Bay, Tinne tuytelaars, and Luc Van Gool,
in Eur. Conf. on Computer Vision (ECCV), 2006
An unbiased detector of curvilinear structures
Carsten Steger,
IEEE Trans. Pattern Anal. Machine Intell.(PAMI) 1996
2009-07-01
46
Outline
Introduction
Detector
Descriptor
Corners and blobs
Curvilinear structures
Orientation
Descriptor components
Validation
Conclusion
2009-07-01
47
Dermoscopic features
A Pigmented Skin Lesion (PSL) typically has several
dermoscopic features
Over 100 of these features
2009-07-01
48
Detecting line points
Cross section
L’ = 0
L’’ large
L(x)
Curve
n(x)
n(x)
2009-07-01
49
[ Steger 1998, ”An Unbiased Detector of Curvilinear Structures”
]
Experiment
2009-07-01
50