Image Segmentation and Searchingusers.monash.edu.au/~dengs/resource/papers/innov02.pdf · vComputer...
Transcript of Image Segmentation and Searchingusers.monash.edu.au/~dengs/resource/papers/innov02.pdf · vComputer...
1
Faculty of Information TechnologyFaculty of Information Technology
Image Segmentation andSearching
Gour Karmakar, Laurence Dooley, M. Manzur MurshedDengsheng Zhang, Guojun Lu
Gippsland School of Computing & Info TechMonash UniversityChurchill, VIC 3842
http://www.gscit.monash.edu.au/
2
Faculty of Information TechnologyFaculty of Information Technology
Image Segmentation
Definition – What is it?vThe process of separating out mutually
exclusive homogeneous regions ofinterest.vThere is no standard formal definition
Definition – What is it?vThe process of separating out mutually
exclusive homogeneous regions ofinterest.vThere is no standard formal definition
Cloud
Urban Scene
3
Faculty of Information TechnologyFaculty of Information Technology
Industry Applications
v Automatic car assembly in robotic visionv Airport security systemsv Object recognitionv Criminal investigationv Computer graphicsvMedical ImagingvMPEG-4 video object (VO) segmentationvMEPG-7 description of multimedia content
4
Faculty of Information TechnologyFaculty of Information Technology
Image
Any SegmentationAlgorithm: 5 Refinement
(FRIS): 2
IncorporatingTexture
(FRIST): 3
Colour
Segmentation
(FRCIS): 4
Evaluation: 6
Numerical
Results
Segmented
Results
Reference
ImageGray Level
Segmentation
(GFRIS): 1
Integrated Fuzzy Rule Based Framework
5
Faculty of Information TechnologyFaculty of Information Technology
Example Segmentation Results (2 regions)Using Different Techniques
a) Cloud Image b) Ref. Regions (c) FCM (d) PCM
e) GFRIS, r=1 f) GFRIS, r=2 g) GFRIS, r=4
R1
R2
6
Faculty of Information TechnologyFaculty of Information Technology
Segmented Results of Our FRIS Algorithm
(a) Food (b) Ref. Regions (c) FCM (d) PCM
(e) GFRIS, r=1 (f) GFRIS, r=2 (g) GFRIS, r=4
R1 R2
R3
R4R5
7
Faculty of Information TechnologyFaculty of Information Technology
Segmented Results of Our FRIS Algorithm
(a) FCM (b) PCM (C) GFRIS r=1 (d) GFRIS r=2 (e) GFRIS r=4
(a) Rocket (b) Ref: (C) FCM (d) PCM (e) GFRIS r=1
(a) GFRIS r=2 (b) FCM (C) PCM (d) GFRIS r=1 (e) GFRIS r=2
R1R2
R3
8
Faculty of Information TechnologyFaculty of Information Technology
Segmented Results of FRIST Algorithm
(a) GFRIS, r=1 (c) FRIST, r=1
(b) GFRIS, r=2 (d) FRIST, r=2
(a) GFRIS,r=1
(b) FRIST,r=1
(c) GFRIS, r=2
(d) FRIST,r=2
R1
R2
9
Faculty of Information TechnologyFaculty of Information Technology
Fuzzy Rules for Colour Segmentation (FRCIS)
Considers the following color models: HSV, RGB,XYZ, YUV, YIQ, YCBCR , CIELAB and OHTA
(a) Original Cloud (b) FRCIS ( HSV r=1) (c) FRCIS (HSV r=2) (d) FRCIS HSV r=4
(e) FCM (HSV) (f) PCM (HSV)
10
Faculty of Information TechnologyFaculty of Information Technology
Color Image Segmentation Using the HSV Model
(a) Original Cloud (b) Reference Image (b) FRCIS r=1 (c) FRCIS r=2
(d) FRCIS r=4 (e) FCM (f) PCM
R1
R2R3
11
Faculty of Information TechnologyFaculty of Information Technology
Image Searching• How do we find an image or object
from a remote database of images?
• Potential applications– Online Shopping– Internet– Digital Library– Designing applications– Education– Medical diagnoses
• How do we find an image or objectfrom a remote database of images?
• Potential applications– Online Shopping– Internet– Digital Library– Designing applications– Education– Medical diagnoses
12
Faculty of Information TechnologyFaculty of Information Technology
Finding similar images from a database.Finding similar images from a database.Finding similar images from a database.
13
Faculty of Information TechnologyFaculty of Information Technology
How does it work?
• Polar Raster Sampling• Polar Raster Sampling
Polar Grid
Polar image Polar raster sampled image in Cartesian space
14
Faculty of Information TechnologyFaculty of Information Technology
• Binary polar raster sampled shapeimages
• Binary polar raster sampled shapeimages
Polar raster sampling
Polar raster sampling
15
Faculty of Information TechnologyFaculty of Information Technology
Feature Extraction
• 2-D Fourier transform on the polar rastersampled image f(r,θ)
• The N normalized Fourier coefficients areused to represent the shape in database.
• 2-D Fourier transform on the polar rastersampled image f(r,θ)
• The N normalized Fourier coefficients areused to represent the shape in database.
feature={f1, f2, …, fN}
16
Faculty of Information TechnologyFaculty of Information Technology
• Each object in the database isrepresented by the extracted features.
• Each object in the database isrepresented by the extracted features.
Indexing
f1: (a0, a2, a3, …, aN)
f2: (b0,, b1, b2, …, bN)
f3: (c0, c1, c2, …, cN).
.
.
fm: (z0, z1, z2, …, zN)
objectsobjectsobjects
featuresfeaturesfeatures
indexing
indexing
indexing
indexing
.
.
.
17
Faculty of Information TechnologyFaculty of Information Technology
Retrieval
• User submits a query object, the systemcompares it with all the indexed objects in thedatabase, the most similar objects are sent tothe user.
• User submits a query object, the systemcompares it with all the indexed objects in thedatabase, the most similar objects are sent tothe user.
UserUserUser
Query object comparing
comparing
comparing
comparing
Objects
inthe
database
similar
.
.
.
18
Faculty of Information TechnologyFaculty of Information Technology
Query
Sea
rchi
ngR
esul
t
Real Searching
19
Faculty of Information TechnologyFaculty of Information Technology
Searching for Pictures of Sydney Opera House
20
Faculty of Information TechnologyFaculty of Information Technology
Searching for a flower