Aucun titre de diapositive

1
OBJECT-BASED VIDEO CODING USING A DYNAMIC CODING
SCHEME
Marc CHAUMONT1, Stéphane PATEUX2, Henri NICOLAS3
CONTEXT Video object coding
OBJECTIVE Demonstrate the competitivity of
Object Coding versus Non Object coding
SOLUTION A Global scheme with 4 video object
coders (Dynamic Coding) Definition of a common
distortion metric A rate distortion
optimization
SLIDE 2
SLIDE 4
SLIDE 4
2
OUR DYNAMIC CODING SCHEME
Galpin et al. 01 Balter et al. 03
Marquant 00 Pateux et al. 01
Cammas et al. 03
ITU-T Rec. H.264 / ISO/IEC 11496-10, 02 JM5
3
STATE-OF-THE ART CODERS
Sprite coder
3D model-based coder
for GOP 108-119
for GOP 0-90
a sprite image
parametric motions
a 3D mesh
a texture
camera motions
t2D wavelet coder
H264/AVC object based coder
for GOP 0-8
for image 3
a 2D mesh
projected textures
Bloc motion
Bloc textures
4
A COMMON RATE-DISTORTION METRIC
A RATE-DISTORTION OPTIMIZATION
OBJECTIVE
RESOLUTION BY LAGRANGIAN FORMULATION
5
RESULTS foreman CIF 15Hz
Background wlt3D
109 Kb/s
Face H264/AVC
147 Kb/s
6
RESULTS CONCLUSION
Thabor Stairs CIF 25 Hz
Stefan 352x240 15Hz
CONCLUSION A clever rate repartition A
clever coding of each object Better than
H264/AVC at low bit-rate (lt256Kb/s, CIF 15Hz) A
positive item for object coding
approaches PERSPECTIVES Replace time
computational consuming step (curves extraction
RD optimization) by a simple prediction
step. Use other metrics close to Human Visual
System