Watermarking of Polygonal Lines
Click here to load reader
Transcript of Watermarking of Polygonal Lines
AIIA Lab, Department of InformaticsAIIA Lab, Department of InformaticsAristotle University of ThessalonikiAristotle University of Thessaloniki
V. R. Doncel, N. Nikolaidis and I. PitasV. R. Doncel, N. Nikolaidis and I. Pitas
Watermarking of Polygonal Watermarking of Polygonal LinesLines
Department of InformaticsDepartment of InformaticsAristotle University of ThessalonikiAristotle University of Thessaloniki
GREECEGREECE
e-mail: e-mail: {victor,nikos,pitas}@zeus.csd.auth.gr{victor,nikos,pitas}@zeus.csd.auth.gr
VISNETThessaloniki 25 June 2004
AIIA Lab, Department of InformaticsAIIA Lab, Department of InformaticsAristotle University of ThessalonikiAristotle University of Thessaloniki
IntroductionIntroduction
Polygonal line: sequence of vertices defining a polygonPolygonal line: sequence of vertices defining a polygon
Polygonal lines in: GIS data, cartoons, segmented images (from Polygonal lines in: GIS data, cartoons, segmented images (from video), CAD, general vectorial graphicsvideo), CAD, general vectorial graphics
Robust watermark system using Fourier descriptorsRobust watermark system using Fourier descriptors
AIIA Lab, Department of InformaticsAIIA Lab, Department of InformaticsAristotle University of ThessalonikiAristotle University of Thessaloniki
Fourier descriptors: Fourier coefficients of the polygon Fourier descriptors: Fourier coefficients of the polygon considered as a function in the complex planeconsidered as a function in the complex plane Sample: 1. USA 2. USA in the Fourier domainSample: 1. USA 2. USA in the Fourier domain
Watermark Embedding: 1Watermark Embedding: 1
-120 -115 -110 -105 -100 -95 -90 -85 -80 -75 -7015
20
25
30
35
40
45
50
55
60X(i) in the complex plane
-30 -20 -10 0 10 20 30-30
-25
-20
-15
-10
-5
0
5
10
15
20X(k) b> k > a
AIIA Lab, Department of InformaticsAIIA Lab, Department of InformaticsAristotle University of ThessalonikiAristotle University of Thessaloniki
Watermark Embedding: 2Watermark Embedding: 2
Watermark:
Spread spectrum techniques
W(k) A pseudorandom signal, generated with an integer key
W(k) takes values of ±1 randomly, N length
Watermark is multiplicative: |X’(k)|=|X(k)| (1+pW(k)) Watermark is only embedded in medium frequencies
AIIA Lab, Department of InformaticsAIIA Lab, Department of InformaticsAristotle University of ThessalonikiAristotle University of Thessaloniki
Correlation is calculated: C=Correlation is calculated: C=ΣΣ |X’(k)|W(k) |X’(k)|W(k)
Random variable with 0 mean if no key or wrong key providedRandom variable with 0 mean if no key or wrong key provided
Compared against a thresholdCompared against a threshold
For big N, it performs well (central limit theorem applies)For big N, it performs well (central limit theorem applies)
Watermark Detection: CorrelatorWatermark Detection: Correlator
AIIA Lab, Department of InformaticsAIIA Lab, Department of InformaticsAristotle University of ThessalonikiAristotle University of Thessaloniki
Signal Im and Re parts considered to be independent gaussian Signal Im and Re parts considered to be independent gaussian processes: Modulus amplitudes follows a Rayleigh distributionprocesses: Modulus amplitudes follows a Rayleigh distribution
Every sample is expected to have a value according to the Every sample is expected to have a value according to the watermark for that point, different if watermark not present.watermark for that point, different if watermark not present.
Likelihood for every sample is considered. Likelihood for every sample is considered.
Better results than the correlator, but slowerBetter results than the correlator, but slower
Watermark Detection: OptimalWatermark Detection: Optimal
AIIA Lab, Department of InformaticsAIIA Lab, Department of InformaticsAristotle University of ThessalonikiAristotle University of Thessaloniki
Correlator is faster but has higher error probabilityCorrelator is faster but has higher error probability Example for a very small embedding power (0.1)Example for a very small embedding power (0.1) In the ROC shown, correlator in green, optimal in redIn the ROC shown, correlator in green, optimal in red
Watermark Detection: ComparisonWatermark Detection: Comparison
10-3
10-2
10-1
100
10-3
10-2
10-1
100
P false detection
P w
ater
mar
k no
t det
ecte
d
AIIA Lab, Department of InformaticsAIIA Lab, Department of InformaticsAristotle University of ThessalonikiAristotle University of Thessaloniki
Non-idealities happen in real life data.Non-idealities happen in real life data.
Variance is not stationary along the spectrum. Improvements Variance is not stationary along the spectrum. Improvements have to be done in the variance estimation.have to be done in the variance estimation.
Watermark Detection: Watermark Detection: ImprovementsImprovements
AIIA Lab, Department of InformaticsAIIA Lab, Department of InformaticsAristotle University of ThessalonikiAristotle University of Thessaloniki
Reads ESRI’s shapefile format GIS dataReads ESRI’s shapefile format GIS data Extracts polygons and applies/read watermarksExtracts polygons and applies/read watermarks
Practical work: PolyWaterPractical work: PolyWater
AIIA Lab, Department of InformaticsAIIA Lab, Department of InformaticsAristotle University of ThessalonikiAristotle University of Thessaloniki
1. Original and watermarked polygon: very small difference1. Original and watermarked polygon: very small difference 2. Rigth vs. wrong keys test: clear detection2. Rigth vs. wrong keys test: clear detection
Practical work: samplePractical work: sample
AIIA Lab, Department of InformaticsAIIA Lab, Department of InformaticsAristotle University of ThessalonikiAristotle University of Thessaloniki
1. Slight differences are visible when zooming in. 1. Slight differences are visible when zooming in. Practical work: sample (2)Practical work: sample (2)
AIIA Lab, Department of InformaticsAIIA Lab, Department of InformaticsAristotle University of ThessalonikiAristotle University of Thessaloniki
ConclusionConclusion
Watermark for polygons robust against attacksWatermark for polygons robust against attacks
Good performance for N > 1000 pointsGood performance for N > 1000 points
When multiple contours, fusion techniques have to be When multiple contours, fusion techniques have to be developed to avoid mismatching between bordersdeveloped to avoid mismatching between borders