Graphics Reference
In-Depth Information
3.4 StereoscopicVideoCoding
Conventionally, stereoscopic video is captured by two cameras a small distance apart pro-
ducing two viewswhichdiffersomewhatofthesameobject.Therefore,stereoscopicvideo
coding aims to exploit the redundancies present in two adjacent views while removing
the temporal and spatial redundancies. For example, an extended H.264 video codec for
stereoscopic video, exploiting spatial, temporal, disparity and world-line correlation is
described in [48]. Instead of coding the left and right views separately, the coding of
left view and the disparity compensated right image has been an efficient method of
coding. Disparity estimation algorithms are utilized to identify geometric correspondence
of objects in stereo pair [49]. A disparity compensated residual image coding scheme for
3D multimedia applications is described in [50]. The proposed stereoscopic video coding
algorithm in [51] employs an object-based approach in order to overcome the artefacts
caused by block-based disparity estimation approaches. However, the disparity-com-
pensated prediction approach can not be applicable to colour plus depth video as colour
imageandassociateddepthimagehavedifferenttextureinformationandnumberofobjects.
In addition, mixed-resolution video coding for the left and right views is based on the
binocularsuppressiontheorem which is based on the response of the human visual system
for depth perception. This theory suggests that 3D perception is not affected if one of
the view can be in high quality compared to the other view [134]. The high quality
view drives 3D perception by compensating the slight loss in the other view. H.264/AVC
based stereoscopic video codec described in [52] utilizes asymmetric coding of the left
and right images in order to improve the compression efficiency. The results in [52] show
that stereoscopic video can be coded at a rate about 1.2 times the monoscopic video us-
ing spatial and temporal filtering of one image sequence. Moreover, a 3D video system
based on H.264/AVC view coding described in [53] examines the bounds of asymmetric
stereo video compression. The mixed-resolution coding concept can be applied to colour
plus depth video in different perspectives, i.e. to code the colour and depth video based
on their influence towards better perceptual quality. For example, the depth image can be
coded differently compared to the corresponding colour image [130].
Themulti-viewprofile(MVP)oftheMPEG-2videocodingstandardcanbeutilizedincoding
stereoscopic video with Disparity-Compensated Prediction (DCP). This profile is defined
basedontheTemporalScalability(TS)modeofMPEG-2 [54]. However,duetothecoarser
disparity vectors which are defined on a block-by-block basis of 16x16 pixels, the inter-
view prediction error is larger compared to the motion compensated error [55]. Therefore,
thegainobtainedwiththiscodingapproachisnotsignificantcomparedtothecodingofleft
and right views separately. However, the proposed improvements (e.g. globaldisplacement
Search WWH ::




Custom Search