Upload
king
View
39
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Une mesure de texture géométrique pour cacher les artefacts en compression et tatouage 3D. Guillaume Lavoué. GDR ISIS – Thème D – Compression d'Objets 3D Statiques et Animés – 2 Avril 2009. Many processing operations on 3D objects. Simplification. Compression. Watermarking. - PowerPoint PPT Presentation
Citation preview
[email protected] - http://liris.cnrs.fr/glavoue
Laboratoire d'InfoRmatique en Image et Systèmes d'informationLIRIS UMR 5205 CNRS/INSA de Lyon/Université Claude Bernard Lyon 1/Université Lumière Lyon 2/Ecole Centrale de Lyon
Université Claude Bernard Lyon 1, bâtiment Nautibus43, boulevard du 11 novembre 1918 — F-69622 Villeurbanne cedex
http://liris.cnrs.fr
UMR 5205
Une mesure de texture géométrique pour cacher les artefacts en compression et
tatouage 3D
Guillaume Lavoué
GDR ISIS – Thème D – Compression d'Objets 3D Statiques et Animés – 2 Avril 2009
2
Many processing operations on 3D objects
SimplificationSimplification
CompressionCompression
WatermarkinWatermarkingg
Distorted objectsDistorted objects
These processes must concerve the These processes must concerve the visual aspectvisual aspect of the of the models.models.
Classic geometric distances Classic geometric distances do not correlatedo not correlate with the with the human visual perception human visual perception
3
Masking and Roughness concepts
Our objective is to exploit some perceptual aspects to hide Our objective is to exploit some perceptual aspects to hide degradations produced by standard operations.degradations produced by standard operations.
This idea is linked with the concept of This idea is linked with the concept of MaskingMasking : : A rough A rough region is able to hide some geometric distorsion with similar region is able to hide some geometric distorsion with similar frequencies.frequencies.
In Computer Graphics In Computer Graphics MaskingMasking was investigated by was investigated by Ferwerda et al. 1997 Ferwerda et al. 1997 Complex computational masking Complex computational masking model.model.
Our objective: Our objective: A simple roughness estimatorA simple roughness estimator, allowing to , allowing to concentrate the distorsion of common operations on concentrate the distorsion of common operations on noisednoised areas associated with high masking levels.areas associated with high masking levels.
4
Outline
Introduction
The proposed roughness measure
Results and application to masking
Integration to compression / watermarking
5
Overview
Two main constraints:Two main constraints:
Our measure has to be Our measure has to be Multi-Scale and independent of Multi-Scale and independent of the mesh connectivity.the mesh connectivity.
EdgeEdge and and smoothsmooth regions have to be clearly regions have to be clearly differentiated from differentiated from roughrough regions. regions.
6
Overview
Over local windows
7
Discrete Curvature calculation Geometric information Geometric information is not relatedis not related to perception to perception
Curvature variations strongly reflect the variations of the Curvature variations strongly reflect the variations of the intensity image after rendering.intensity image after rendering.
[Cohen-Steiner and J. Morvan, 2003]Restricted delaunay triangulations and normal cycle
Curvature tensor at each vertex of the meshCurvature tensor at each vertex of the mesh
EigenvaluesEigenvalues Principal curvature values Principal curvature values KminKmin, , KmaxKmax
2ii
i
KminKmaxvC
)(
8
Curvature averaging
9
Adaptive smoothing Main problem with classical smoothing (Laplacian):Main problem with classical smoothing (Laplacian):
Our adaptive smoothingOur adaptive smoothing Derived from the two-step filter Derived from the two-step filter [Taubin, 1995][Taubin, 1995]
Dependent of the sampling density Independent of the sampling density
10
The roughness measure
1.1. The 3D object is smoothed (The 3D object is smoothed (εε scale window) scale window)
2.2. Curvature is calculated for both meshes (original Curvature is calculated for both meshes (original and smoothed)and smoothed)
3.3. Average curvature is processed for each vertex Average curvature is processed for each vertex ((22εε scale window)scale window)
4.4. Asymmetric curvature difference for each vertex Asymmetric curvature difference for each vertex Roughness map Roughness map
11
Outline
Introduction
The proposed roughness measure
Results and application to masking
Integration to compression / watermarking
12
Results
ε = 1 % ε = 3 %
13
Comparison
14
Robustness to connectivity change
Sam
pling density
15
Application to Masking
Original Two clustersRough / Smooth
Noise on smooth regions
Noise on rough regions
Same RMS distance
Much more visibleMSDM = 0,42
MSDM = 0,36
Rough regions exhibit a higher masking degree.Rough regions exhibit a higher masking degree. Distorsion errors coming from common processing operations Distorsion errors coming from common processing operations
can be concentrated on these areas.can be concentrated on these areas.
16
Subjective experimentThe 3D corpus 4 objects 6 versions : 3 noise strengths
on smooth and rough areas
Evaluation protocol 6 degraded versions are displayed to the observer together with
the original object He must provide a score between 4 (identical to the original) and 0
(worst case)
Results
17
Outline
Introduction
The proposed roughness measure
Results and application to masking
Integration to compression / watermarking
18
Integration to single rate compression
Roughness analysis
19
The algorithm
Connectivity coding Face Fixer [Isenburg and Snoeyink 2001]
Geometry coding Simple differential coding Variable quantization: lower for rough region, higher for smooth ones
Arithmetic coding
Roughness classification Markov based clustering [Lavoué and Wolf 2008]
20
Results
21
Integration to spectral watermarking
The Ohbuchi et al. [2002] non blind scheme:
Mesh segmentation into patches
Spectral decomposition of each patch
Modulation of spectral coefficients (fixed strength α)
Non blind extraction
Watermark WatermarkWatermark
Roughness analysis
22
Illustration
Roughness map Segmented regions
Adaptation of the VSA [Cohen-Steiner et al. 2004]
23
Visual results
Original Ohbuchi et al; 2002 Ours
24
Robustness
2 attacks Noise addition Non uniform scaling 50 insertion / extraction
25
Conclusion
Un algorithme de caractérisation de la rugosité de la surfaceMise en évidence du phénomène de Masking par une expérience subjectiveRésultats encourageants après intégration pour la compression et le tatouagePour + d’info: Lavoué, G. 2009. A local roughness measure for 3D meshes and its application to visual masking. ACM Trans. Appl. Percept. 5, 4 (Jan. 2009), 1-23.
Et maintenant : Caractérisation plus théorique du phénomène par des expériences subjectives plus poussées