Graphics Reference
In-Depth Information
The MPEG-4 Animation Framework eXtension (AFX) standard ISO/IEC 14496-16
defines two image- based depth representations that can be used to encode colour plus
depth video
[58].
The two categories are the simple Depth Image (DI) and Layered Depth
Image (LDI) representation which is utilized to store occluded areas of the original image.
However, this approach seems very much computer graphics oriented and suitable for
syntheticimagesynthesis
[58].
Therefore,thisapproachisnotsuitableforencodingnatural
scenes using monoscopic video plus depth representations of 3D video.
The quality of the virtual left and right views generated through DIBR is dependent on the
qualityof thecolourvideo,associateddepthmapandthecamerageometry.Therelationship
canbeattributedto theQuantizationParameters(QPs)ofeachimageandthevirtualcamera
position(seeEquation3.1).
Quality =
f
(QP
colour
, QP
depth
, Camera position)
Equation 3.1
However, the effect of colour image quality is more influential with this approach than the
depth image quality as texture information is directly viewed by the users. Therefore, ef-
fects of colour and depth coding on the perceptual quality needs to be studied thoroughly.
Thischapterinvestigatestheperceptualboundsofdepthmapcodingforstereoscopicvideo
applications. For example, spatially down-sampled depth maps would be accurate enough
to generate good quality left and right views. Moreover, the efficient coding architectures
for colour plus depth map coding and their suitability for communication applications are
reported in this chapter.
Due to the characteristics of the depth image sequence (e.g. smoothness of real world
objects),itcanbeefficientlycompressedwiththeexistingcodingalgorithmsthanthecoding
performsMPEG-2andMPEG-4videocodingstandardsindepthmapcoding.Forexample,
depth maps can be encoded with H.264/AVC at a bitrate less than 20% of the MPEG-2
codedcolourimagesequenceforStandardDefinition(SD)resolutionimage sequences. The
followingsubsectiondescribespotentialencodingapproachesforstereoscopic3Dvideo.
3.4.1
ExplorationofEfficientStereoscopicVideoCodingConfigurations
