[ Pobierz całość w formacie PDF ]
.In this case the decoderrequests one of several correctly received and decoded frames to beused as the reference in intercoded pictures instead of an erroneousone.This requires considerable codec memory.• Video redundancy coding (VRC): This is another option that canbe implemented in conjunction with reference picture selection forerror-prone PS networks.Instead of basing sequences of P-frames ona unique reference picture, multiple threads can be defined, each ofwhich is based on a separate reference picture.The effect of an errorin one reference picture is then confined to only one thread.Periodi-cally a common synchronization frame is sent on all threads.H.263 uses several H.245 features for support of video-feedback asfollows:280QoS in Integrated 3G Networks• Freeze frame: picture frozen until release or a 6-second time-out;• Fast update request: request for intracoding of macroblocks for errorlimitation;• Freeze frame release: back to normal transmission after either of thetwo preceding options;• Continuous presence multipoint: a multipoint control unit that cantransmit up to four independent video streams for split-screen pres-entation in a single H.263 video stream.H.263+ offers three scalability options: spatial, temporal, and SNR.Spatialscalability typically involves two sets of pictures, basic and enhanced, with the latter sent at twice the resolution in each plane (e.g., CIF instead ofQCIF).Temporal is given by two frame rates offered by either inclusion orexclusion of B-frames.SNR is similar to that for MPEG with an optional setof quantization errors.In an IP environment, distinct IP addresses and portnumbers are used for the different options so that a user can easily log on to the appropriate quality.H.263+ pictures are transmitted using a multiple layer structure withdifferent headers for each layer as shown in Figure 10.5 [36].The meanings of the uninterpreted acronyms in this diagram are asfollows:• CPM stands for continuous presence multipoint and is a flag thatindicates that four independent video bit-streams may be sentLayerHeader fields with sizes (bits)PictureStartTempPicturePict (5)CPMPEIcode (22)ref (8)type (13)quantzn(1)(1)StartGOBGOB IDGOBGroup of blockscode (17) nom (5)(2)quantznGBSICBPYDQUANTCoded motionMacro blockCOD (1)MCBPC(var.)(var.)(2)vectorsINTRADCBlock(8)TCOEF (coded DCT coefficients)Figure 10.5 H.263 frame layers.Specific Applications281within a single H.263 video channel—if so, the field GBSI is used inthe GOB header.• PEI is a flag that indicates the presence of an optional field.• GBSI is the GOB substance indicator for the streams in the mul-tipoint case indicated by the CPM flag.• The COD flag is used to show whether the macroblock has beencoded, or skipped because it has not changed significantly.• MBCPY is a variable length field that describes the type of mac-roblock and which chrominance blocks are coded.• The CBPY field shows which of the four luminance blocks withinthe macroblock have been coded (some may have been skipped forlack of change).• DQUANT shows the change in quantization parameters for themacroblock since the previous one, and for the first macroblock isset to the default quantization parameter (QP).• The block layer shows the coding of the actual video data, whereINTRADC represents the constant coefficient in the DCT expan-sion for an I-frame, and TCOEFs describe the coefficients forthe cosine terms.Some operations, such as fast-forward, only useINTRADC.If slice structure mode is used in H.263+, then a slice structure header is used instead of the GOB header.This includes a specification of the slice size in addition to comparable material to that for the GOB.H.263+ also uses resynchronization markers and motion boundarymarkers for data partitioning as shown in Figure 10.6.This shows how the macroblock is split into parts for motion vectorsand picture data.The number of macroblocks following the RM is provided,followed by the default QP, followed in turn by the motion data for eachsuccessive macroblock.After the MBM, the data headers for each mac-roblock are given, with the actual DCT data for the individual macroblocksbehind the set of headers.H.263 version 3 (or H.263L) is intended to provide still better per-formance on low bit rate links and on poor quality networks.It is alsointended to have a much greater compatibility with MPEG-4.A further improvement to H.263 is the recent addition of profiles toclassify sets of functionality (H.263 Annex X) for version 2.Those most relevant to UMTS are as follows:282QoS in Integrated 3G NetworksUse of resynchronization marker and motion boundary markerRM MB#QPMotion dataMBMDCT datawhere the form of motion data and DCT data for H.263 is:CODMCBPCMotionCODMCBPCMotionvectorsvectorsCOD.CBPYDQUANTCBPYDQUANT.DCT dataDCT data.Figure 10.6 Data partitioning.• The baseline profile (profile 0) stands for H.263 with no optionalmodes of operation.It includes the basic coding tool set common inmodern video coding standards [ Pobierz całość w formacie PDF ]

  • zanotowane.pl
  • doc.pisz.pl
  • pdf.pisz.pl
  • funlifepok.htw.pl

    •