DR.

KRUPA RASANE
JAIN COLLEGE OF ENGINEERING, BELAGAVI
EMAIL: [Link]@[Link]
MOBILE: 9341107810

1
Krupa Rasane
 MULTIMEDIA
COMMUNICATIONS
MODULE 4
AUDIO & VIDEO COMPRESSION
BEC613A

2
Krupa Rasane
SYLLABUS
3

KRUPA RASANE
Multimedia Communications
Module 4 (Part-2)
VIDEO Compression
BEC613A

Dr. Krupa Rasane 4

Video Compression
• Multimedia application-Interpersonal, interactive and entertainment
• Different digitization formats
• The bitrates required for digi formats are higher than the ones
available in the transmission networks
• Hence compression needs to be applied.
• Topics:
• Video Compression principles, H.261, H.263, MPEG, MPEG-1, MPEG-
2, MPEG-4

Dr. Krupa Rasane 5

Video Compression principles
• Moving pictures- video is simply a sequence of digitized pictures.
• Moving JPEG (MJPEG)-One approach to compressing a video source is to apply the JPEG algorithm
to each
frame independently. Compression ratio-10:1 to 20:1 – not sufficient
• Spatial redundancy is seen in a frame as well as among frames.
• Movies have less changes in consecutive frames – eye movement/lip movement (video
telephony appln), In
movies, movement of vehicles or people
• Typical scene in a movie may have a mini duration of 3 seconds.
• With RR of 60 frames /sec, each scene is composed of 180 frames
• Additional savings in BW can be made by exploiting the temporal differences that exist
between many of the frames.- by sending information related to only the frames that involve
movements.
• Predictive frames are also involved. Some of the transmitted frames may be frames with original
data, some with difference between the actual and the predicted
• Motion Estimation: Prediction is based on the correctness of the estimation done
• Motion Compensation: Additional info sometimes may be required to indi the diff between
predicted and
actual positions of the moving segment. Dr. Krupa Rasane 6
Video Compression principles-Frame types
• Two basic types:
• Intracoded frames(I frames) -ones encoded independently-
• Predicted –intercoded frames: Two types: Predictive (P Frames) and Bidirectional (B frames)
• Typical sequence involes only I and P frames
• I Frames: Y, Cb, Cr are encoded independently using the JPEG algorithm – DCT, Quantization and entropy
encoding except that quan threshold values that are used are the same for all DCT coefficients.
• Level of compression of I frames are relatively small. Hence generally they are limited to say the first frame
relating to a new scene.
• Compression algo is indep of the contents of frames and hence has no knowledge to the start and end of the
scenes.
• I-frames must be present in the output stream at regular intervals to deal with corrupted frames.
• If an I frame is corrupted, the complete scene is lost. So they are inserted rela frequently in the ouput stream.
• The number of frames / pictures between succ I – frames is known as a group of pictures or GOP. Symbol is N
and value between 3 and 12.
• Encoding of P frame is rela to preceding P frame or I frame
• P frames- high levels of compression as motion estimation and compensation are involved
• No of P frames in between every I frame is reduced to avoid propagation of errors
• Prediction span- No of P frames along with immed I frame. Rep by M- range 1 through 3
Dr. Krupa Rasane 7
Dr. Krupa Rasane 8
 Video Compression principles-Motion Estimation
• Motion estimation involves comparison of small segments of two consecutive frames for differences and a
search is carried out to determine to which neigh seg the original has moved. To reduce the time for search,
search is limited to just a few neigh segments. This works well for video telephony as movements are less.
• Fast movements demand search in a wider range. Here B frames are introduced.-involves both past and
future frames. This also is used when an object moves in front of or behind another object.
• With P frames – contents of 2 uncoded frames UF are involved.
• With B frames – 3 UF are involved – prev, curr and next. Introduces a delay- highest level of compression
achieved, do not propagate errors as no frames are coded with these.
• Decoding of I frames immediate. Decoding of P frames involves deco of current and using with that of prev
frame and that of B frames involve 3 frames.
• Order of encoding hence is generally changed. For Ex., If the uncoded frame sequence is IBBPBBPBBI……
then the reordered sequence would be IPBBPBBIBBPBB…
• Fourth type – PB frame – not new – encoding vary- consecu P and B are encoded together to increase the
frame rate without signi increase in bit rate
• Fifty type – D-frame used in movie/ video-on demand-As any part of the movie may be watched, rewind, FF
facilities are available, demands higher decompression speeds, hence highly compressed D frames are used.
These are ignored during the decoding of P and B frames.
• DC coefficient is asso with the 8 X 8 pixel block- both for luminance and two chrominance signals – mean
of all values in the related block. Only the DC coeff of D frames is used and a low resolution seq of frames is
provided each of which can be decoded at higher speeds that ae expected with the rewind and FF
operation. Dr. Krupa Rasane
Nandhini Vineeth 9
6
Dr. Krupa Rasane 10
 Motion estimation and compensation
• The digital Y matrix associated with each frame are first divided into a 2D matrix
of 16 X16 pixels known as macroblock.
• 4:1:1 format- Cb and Cr- 8X8 pixels.
• An address is associated with each macroblock and this consists of four DCT
blocks for luminance and one each for two luminance signals.
• To encode a P frame, each macroblock in this target frame is compared to that of
the previous reference frame. If there is a close match then only address of
macroblock is encoded. If match is not found, then the search is extended to
cover an area around macroblock in ref frame. Search is normally on Y signal.
• If the mean of the absolute errors observed in the frames is below threshold, a
match is said to be found. Two param are encoded. The first is motion vector
indicated by(x,y) offset of the macroblock being encoded and the location of the
block of pixels in the reference frame which produces the match.
• The search and offset can be either on macroblock boundaries or pixel
boundaries. The motion vector is then said to be single pixel resolution.
Dr. Krupa Rasane 11
Motion estimation and compensation
• The second is prediction error- three matrices (one for each Y,Cb,Cr) each of
which contains the diff values between those in target macroblock and the set
of pixels in the search area producing the closest match.
• Physical area of coverage of a macroblock is small, the motion vectors can be
relatively large values.
• Most moving objects are normally larger than a single macroblock. When an
object moves, multiple macroblock are affected.
• Motion vectors are encoded using differential encoding (DE) and the resulting
codewords are then Huffmann encoded.
• The three matrices are encoded- DCT, Quantization and entropy encoding. If a
match cannot be found, the macroblock is encoded independently in the same
way as the macroblocks in an I-frame
• Encoding a B frame any motion is estimated with reference to both imme prece
and succ - I or P frames
Dr. Krupa Rasane 12
Dr. Krupa Rasane 13
• Three sets of Motion vector and diff matrices calculated
• MD - motion estimate done with previous and target frame
• MD’- motion estimate done with succeeding and target frame
• MD’’-Target and mean of the above two
• The set with the lowest set of differences in chosen and encoded the
same way as P frames
• Motion vector is said to be a resolution of a fraction of a pixel. –Half
pixel resolution

Dr. Krupa Rasane 14

Dr. Krupa Rasane 15
Implementation Issues
• In case of P frames , the encoding of each macroblock is dependent
on output of the motion estimation which depends on contents of
MB being encoded and the contents of the macro block in the
search area of the reference frame that produces the closest match
to that being encoded.
• There are three possibilities:
• If the two contents are the same, only the address of the macroblock
in the ref frame is encoded
• If the two contents are very close, both the motion vector and
the diff matrices associated with the MB (macro block ) in the ref
frame are encoded
• If no match is found, then the target MB is encoded in the same
way as a macroblock in an I-frame.
The motion estimation unit containing the search logic, utilizes a
copy of the uncoded reference frame.
Dr. Krupa Rasane 16
• Performance
• Compression Ratio – I frames- 10:1 and 20:1
• Compression Ratio – P frames- 20:1 and 30:1
• Compression Ratio – B frames- 30:1 and 50:1

Dr. Krupa Rasane 17

H.261
• This Video compression standard has been defined by the ITU-T for video
telephony and video conferencing services over ISDN
• Px64 (or kbit/s) is the common nickname for the ITU-T H.261 video coding
standard. It was designed for real-time video conferencing over ISDN lines,
where the data rate (p) is a multiple of 64 kbit/s, typically p->1 to 30
• CIF : Crystallographic Information File
• Digital format is either CIF (Video conferencing) / QCIF (Video Telephony)
• Each frame is divided into macroblocks of 16 X 16 pixels for compression.
• As both use subsampling, the spatial resolution of each format is
• CIF- Y=352 X 288 Cb=Cr=176 X 144
• QCIF- Y=176 X 144 Cb=Cr=88 X 72
• Frame RR = 30fps for CIF AND either 15 or 7.5 fps for QCIF
• only I and P frames used with three Dr.
P Krupa
frames
Rasane between each pair of frames. 18
Dr. Krupa Rasane 19
H.261 encoding formats- MB format,
Frame/Picture format, GOB format
• Type- Intra coded or Intercoded
• Address- Address associated with MB
• Quantization Value- Threshold value used to quantize all DCT coefficient in MB
• Motion vector- encoded vector uses 16x16 Macro Block of 4:1:1 for Y CrCb.
• Coded block pattern – which of the 8X8 pixel blocks that make up the macroblock
are present- for those present, the JPEG encoded DCT coeff are given in each
block
• Picture start code- start of each new video frame/picture is indicated by PSC
• Temporal reference- Timestamp used by RX to sync audio and video of the scene
• Picture Type- I or P frame
• Group of blocks (GoB)- encoding happens on each MB, the data structure
showing a group of MB is GoB

Dr. Krupa Rasane 20

• GOB- 11 x3 Macroblock- size chosen so that both CIF and QCIF contains integral
number of GOBs.-12 for CIF and 3 for QCIF (Interworking b/w formats allowed)
• GOB start code – at head of each GOB-unique start code (var length codeword) –
no entropy encoding codeword produces the same code. Rx detects a corruption,
it starts looking for the start of next GOB and hence said to be resynchronization
marker.
• Group number- every GOB is associated with a group number which allows the
string of GOBs to be missing from a frame. This is observed when the amount of
data observed is more than the BW availability
• FIFO Buffer-
• used to convert a variable bit rate stream into a constant bit rate for the optimal use of BW.
• To reduce the fluctuation of availability of data in FIFO Buffer, sometimes may go into overflow
sometimes go empty, 2 threshold values- low threshold and high threshold values are defined.
• Quantization threshold is modified to control the input data towards FIFO buffer based
on the two thresholds
• This control procedure act in the GOB level rather than MB level. GOB are sometimes
dropped until the overload subsides.

• Dequantizer in Rx informed accordingly.

• Complete frames also get dropped someDrt. iNm
andehs
inid
Viu
nee
ethto BW in availability.
Dr. Krupa Rasane 182
1
Dr. Krupa Rasane 22
H.263
• Video compression std used for wireless video applications and
PSTN.
• Applns- video telephony, video conferencing, security surveillance,
interactive games playing, etc., - real time
• PSTN – works in analog hence require a modem
• Bit rates vary from 28.2 kbps through 56 kbps.
• Video encoder requires the compression to very low bit rates
• Basic std used are as H.261.
• Blocking artifacts- high quantization threshold values lead to this
• Jerky movements may result due to low bit rate. To avoid this some
advanced coding options compared to h.261 used.
Dr. Krupa Rasane 23
H.263 – Digitization formats
• As subsampling is used , the two alternative spatial resolutions are
• CIF- Y=352 X 288 Cb=Cr=176 X 144
• QCIF- Y=176 X 144 Cb=Cr=88 X 72
• Decoder needs the implementation of both but is not expensive as
motion estimation is not seen
• In encoder, game playing applications use S-QCIF and video
conferencing encoders use CIF
• All three I, P and B frames used to obtain higher levels of
compression. PB frames also used sometimes. Encoder : B then P
Decoder: P then B

Dr. Krupa Rasane 24

H.263
• Unrestricted motion vectors
• IN H.261 motion vectors associated with predicted macroblocks are restricted to either edge of the frame or to a defined
area around loc in target frame. This leads to intracoded encoding.
• Here in H.263, unrestricted motion vector mode is used which allows to match even outside the frame boundaries
• Error resilence
• H.263 – is used in PSTN, related high prob of transmission errors.
• A set of error free frames will be followed by burst error affecting a string of macroblocks within a frame.
Error detection in GOB is easier compared to individual macro blocks
• As there is a dependability among frames, errors propagate fast in GOB making it apparent to the viewers.
• When an error is detected in GOB, decoder skips the GOB and moves to decoding next by searching for sync
marker at the head of a GOB.
• ERROR CONCEALMENT SCHEME- repetition of a previous frame is seen
• As low bit rate is available in PSTN, I frames are seen infrequently. When QCIF, results are very annoying.
• Figure shows the propagation of errors- initial and to the various depended frames
• Schemes –error tracking, independent segment decoding and reference picture selection are used to
minimize the effect of errors.

Dr. Krupa Rasane 25

H.263-Error Tracking
• With real world applications such as video telephony as a two way communication is
available, the decoder uses the reverse channel to indicate the encoder that an error has
occurred using a NEG ACK-combn of frame number and loc of GOB in the frame.
• Errors are detected in a number of ways:
• One or more out of range motion vectors
• One or more invalid var-length codewords
• One or more out of range DCT coeff
• An excessive number of coefficients within a macro block
• The encoder retains an error prediction information of all GOBs in each of the most
recently transmitted frames- likely spatial and temporal effects on the macro blocks in
the foll frames that will result if a specific GOB frame is corrupted.
• Using the error prediction info the encoder then iden the macroblocks where errors
would have occurred. Sends the macroblocks in intracoded form(I frames). NAK-(1,3)
indicate error in frame 1- GOB 3. In the next frame 4, the affected macroblocks are
intracoded rather than predicted.

Dr. Krupa Rasane 26

Dr. Krupa Rasane 27
H.263- Independent Segment Decoding
• To overcome errors with in a GOB affecting the neighbouring GOBs in succeeding frames,
this is introduced.

• Each GOB is considered to be a subvideo which is indep of the other GOBs in the frame.
Therefore, the motion estimation and compensation is limited to the boundary pixels of
a GOB rather than a frame.

• When a GOB of a frame is affected, the same GOB in each succ frame is affected until a
new intracoded GOB is sent by the encoder – neighbouring GOBs are not affected.
• Limitation- efficiency of the motion estimation and compensation in the vertical
direction is reduced signi owing to the search area being limi to a single GOB

• This is not use as a standalone but in conjunction with either the error tracking scheme
or more usually with the ref pic scheme

Dr. Krupa Rasane 28

Dr. Krupa Rasane 29
H.263- Reference Picture Selection
• Simi to ET (Error Tracking) in that it tries to stop errors propagating by the decoder returning ack
message when an error in GOB is detected
• The scheme can be operated in two diff modes
• 1. NAK mode:
• only GOBs in error are signaled by the decoder returning a NAK message.
• Intercoded frames depend on Intracoded frames.
• A copy of decoded frame is retained by the encoder. Any of the previous decoded frames can be used as the ref.
In fig shown, when the NAK(2,3) related to frame 2 is received, the encoder selects GOB 3 of frame 1 as the ref to
encode GOB3 of the next frame –frame 5.
• With this scheme, GOB in error will propagate for a number of frames, the number being determined by the
round trip delay of the communications channel, NAK sent and an I frame received
• 2. ACK mode:
• All frames without errors are acknowledged by the decoder with an ACK message.
• Only ack frames are used as ref.
• As seen in figure, lack of ACK for frame 3, makes encoder use frame 2 for frame 6.
• When ACK for frame 4 is received, it is used to encode frame 7.
• Encoding efficiency of the frame is reduced because of the usage of a distant reference frame
• ACK mode performs best when round trip delay of the communications channel is short and less than the time
the encoder takes to encode each frame
•
Dr. Krupa Rasane 30
Dr. Krupa Rasane 31
Dr. Krupa Rasane 3
2
MPEG- Motion Pictures Expert Group
• MPEG was formed by ISO to formulate a set of standards relating to a
range of multimedia applications that involve the use of video with
sound.
• Three standards related to either recording or transmission of the
integrated audio and video steam, each targeted at a particular
application domain. They are
• MPEG-1
• Subsets of ISO Reco – 11172.
• Video resolution is based on SIF (Source Intermediate Format)-resolution-352 X288
pixels
• Intended for storage of VHS- quality audio and video on CD-ROM and bit rates –
1.5Mbps. Higher bit rates are also common to provide faster access to stored material

Dr. Krupa Rasane 33

MPEG
• MPEG-2
• Subsets of ISO Reco – 13818
• Intended for recording and transmission of studio quality audio and video
• Four standards:
• Low-
• Video resolution is based on SIF (Source Intermediate Format)-resolution-352 X 288 pixels
• Intended for storage of VHS- quality audio and video on CD-ROM and bit rates – 4Mbps.
• Main –
• Based on 4:2:0 digi format with a resolution of 720 X 576 pixels
• Produces studio quality digital audio and video
• Target Bit rate is upto 15Mbps or 20Mbps with 4:2:2 digi formats
• High 1440:
• based on 4:2:0 digi format with a resolution of 1440 X 1152 pixels
• Intended for high-definition Television(HDTV)- Bit rates upto 60 Mbps or 80 Mbps with 4:2:2 format
• High:
• based on 4:2:0 digi format with a resolution of 1920 X 1152 pixels
• Intended for wide screen high-definition Television(HDTV)- Bit rates upto 80 Mbps or 100 Mbps with 4:2:2
format

Dr. Krupa Rasane 34

Dr. Krupa Rasane 35
MPEG
• MPEG-3 standard focused on HDTV was not developed separately but
merged with MPEG-2
• MPEG -4:
• Initially was considered with similar range of applications to those of H.263,
each running over very low bit rate channels ranging from 4.8 to 64kbps
• Later expanded to a wide range of interactive MM applications over the
Internet and the various types of entertainment networks
• MPEG-7:
• Concerned with describing the structure and features of the content of MM
info produced by the diff stds.
• Resulting descriptions are used in search engines to locate particular items of
material that have a defined feature. Not discussed here.
Dr. Krupa Rasane 36
MPEG
• MPEG-1,MPEG-2 and MPEG-4 are in three parts: video, audio and system. The first two are
concerned with the way in which they are compressed. The system is concerned with how
the two streams are integrated together to produce a synch o/p stream.
• MPEG-1
• Uses a similar video compression technique as H.261.
• Each frame is divided into 16 X 16 pixels
• Hori resol reduced from 360 to 352 (integ number of 22 macroblocks)
• Since the two chrominance signals are subsampled at half the rate of the lumi signal, the spatial
resolution for the two types of video sources are:
• NTSC: Y=352 X 240, Cb=Cr=176 X 120
• PAL: Y=352 X 288, Cb=Cr=176 X 144

• Appln domain – storage, progressive scanning used RR- 30 Hz(NTSC) and 25 Hz(PAL)

• The standard supports – I , P and B not D

• I frames used for various random access functions asso with VCRs. Accepted max random access is
0.5 sec and so this is the main factor- along with video quality that influences the max sep of I frames
in the frame sequence .
• Original sequence of PAL – IBBPBBPBBI…. and NTSC – IBBPBBPBBPBBI….
• The second (NTSC) sequence is used in both systems.
Dr. Krupa Rasane 37
MPEG-1
• The compression algo is based on H.261
• Each macroblock (MB) of 16 X 16 in Y plane and 8 X 8 pixels in Cb and Cr
planes. There are two main differences:
• 1. Timestamp- Temporal references can be inserted within a frame to
enable the decoder to resynchronize in case of corrupted or missing
frames. The number of macroblocks between two Timestamps is said to be
a slice. [1 .. Max no of MB in a frame] Typically 22- no of MBs in a line
• [Link] of B frames which introduces a time interval between I and
P frames. Search window in the reference is increased. To increase the
accuracy of the motion vectors, a finer resolution is used.
• Compre ratio – I frame—10:1, P frames-20:1, B frames – 50:1

Dr. Krupa Rasane 38

MPEG -1
• Complete compressed video is said to be a sequence consisting of a
group of pictures(GOP)
• Each picture is made up of N slices. For the decoder to decompress
the rx stream, each data structure must be clearly identified within
the bit stream.

Dr. Krupa Rasane 39

Dr. Krupa Rasane 40
Format of MPEG -1 video bit stream
Bit rate
Screen
and size of Contents of Quantization
size and
memory/ tables that are used for
aspect
Start of Frame various frames/picture types
ratio
the buffers
sequence
Pictures
are
seperated
GOPs are
separated I/B/P
using this
How full memory
TS- needs to be before
Synchronization decoding starts
Sequence
Resolution used
of Frame
for motion vectors
types
Scan line Scaling factor
Slices are theDDr.
[Link]
caendhin
Rasane 374
tihVainteaetphplies to
seperated relates to 1
MPEG-2
• There are four level targ a sep appln domain – low, main, high 1440 and
high
• There are 5 profiles with each level- simple, main, spatial resolution,
quantization accuracy and high
• This 4 X 5 two dimen table acts as a framework for all std activities.
• This was targeted to develop the existing standard and intro new standards
rela to one [Link]. At a particular level, the decoders used with a given
profile will be able to decode all the lower profiles that have been defined
for that level. This leads to the internetworking of older and newer
equipments.
• First level is compatible with MPEG1.
• Rest three levels are discussed

Dr. Krupa Rasane 42

 MP@ML
• Target appln in digi TV Bx.
• Interlaced scanning used, RR-30Hz or 25Hz.
• 4:2:0 digi format Reso: 720 X 480 pixels (30Hz)/725 X 576(25Hz)
• Output bit rate is 4 Mbps to 15 Mbps (dep on BW avai in bx channel)
• Video coding scheme same as MPEG-1 Except this uses interlaced.
• As alternative lines are present in each field, the ques is how the DCT blocks are derived
from each macroblock.
• Fig b and Fig C show the two alternatives possible depending on whether the DCT Blocks
are derived from the lines in the field mode or the lines in the frame mode.
• Frame refresh rate -30/25 Hz, field refresh rate - 60/50 Hz
• The amount of motion present in the video decides the frame of field mode
• If large- DCT enc operation on the lines in a field as a higher compression ratio is produced
as the succ fields have shorter time interval
• If little movement is observed, frame mode is used since the longer time interval between
succ frames is less important

Dr. Krupa Rasane 43

MP@ML
• The macroblocks/ DCT blocks are derived from the lines in ech complete frame
• EX. Live sport event is encoded using field mode and a studio based program the frame mode
• Three modes – field, frame and mixed are possible – motion estimation asso with the encoding of
macroblocks in P and B frames
• In field mode, for both P and B frames, the motion vector for each MB is computed using the
search window around the corresponding MB in imme prec field
• For B frames the imme succeeding P/I field.
• The motion vector relate to the amount of movement htat has taken place in the time to scan one
field.
• In the frame mode, a MB in an odd field is enco rel to that in prec/succ odd fields and simi for MB
in even fields. Here motion vector relate to amt of movement that has taken place in te time to
scan two fields- time to scan a complete frame.
• In the mixed mode, the motion vectors for both field and frmae modes are computed and the one
with the smallest value is selected

Dr. Krupa Rasane 44

HDTV
• There are three standards asso with HDTV: adv TV(ATV) in North Amer,
Digital Video Broadcast in Europe, Multiple sub-Nyquist sampling
encoding(MUSE) in Japan and the rest of Asia.
• In addition, to giving the digi formats and compression schemes used also
give info on resulting bit streams transmitted over the diff types of
broadcast n/w.
• As with normal digital TV, there is an ITU-R HDTV specification concerned
with the stds to be used in TV studios for the production of HDTV programs
and also for the international exchange of programs
• Defines 16/9 aspect ratio with 1920 samples per line and 1152(1080
visible) lines per frame
• Interlaced scanning is used with 4:2:0 digi format
• Future prog scanning may be intro for 4:2:2
Dr. Krupa Rasane 45
Dr. Krupa Rasane 46
• Grand Alliance- ATV in assoc with large number of manufacturers
include ITU-R HDTV as well lower resolution format
• 16/9 Aspect ratio resolution- 1280 X 720
• Compr algo is based on MP@HL
• DVB –
• 4/3 AR Resol- 720 X 576
• Video compr algo is based onSSP@H1440 spatially scaleable profile at high
1440 OF MPEG 2 which is similar to MP@HL
• MUSE
• 16/9 1920 x 1035
• Compr simi to MP@HL

Dr. Krupa Rasane 47

MPEG-4
• Appln domain of MPEG-4 is the audio and video associated with interactive
multimedia applications over the Internet and various types of
entertainment networks.
• Content based functionalities
• Audio visual objects (AVO)
• Combn of audio and video components
• Object descriptors
• BIFS- BINARY FORMAT FOR SCENES
• Consists of commands for video objects – change of color/texture/shape
• Consists of commands for audio objects- Voice amplitude/depth
• Scene descriptors –position of each AVO in the screen
• Subobjects exists- person –divided to head/eyes/mouth

Dr. Krupa Rasane 48

• Each frame is segmented into a number of video object planes each
of which corresponds to an AVO
• 3 VOPS IN THE FIGURE
• Spatial coordinators are with respect of top left mb

• Audio and video compression

• Transmission format

Dr. Krupa Rasane 49

Dr. Krupa Rasane 50
Dr. Krupa Rasane 51
Dr. Krupa Rasane 52
Error Resilence techniques
• Important with wireless networks and PSTN:
• USE OF FIXED LENGTH VIDEO PACKETS AS BASE LEVEL DS IN PLACE OF GOBS
• A new variale length coding scheme based on reversible VLCs

• Video Packets:
• Resync markers
• Level of compression depends on level of activity
• As against the equal number of macroblocks in other standards, MPEG uses equal
number of bits. The resulting group of bits is known as video packet and these are
sep by resyn markers
• Resync markers are increased in the area where more activity is seen
• When an error occurs, only limi no of MBs are affected
• Sync markers posi depends on the link bit rate of the transm channel

Dr. Krupa Rasane 53

Dr. Krupa Rasane 54
• VLC- Variable length codewords
• Any bit error cause the decoder to lose sync and discard the
remaining codewords until the next motion boundary or resy marker.
• To over come this reversible VLC are used
• This can be encoded when read from either l to r or right to left
• This helps for both fwd or reverse direction decoding
• RVLCs are produced using constant Hamming weight- each codeword
has same number of binary 1s
• Fixed length prefix and suffix is added to each of the VLC

Dr. Krupa Rasane 55

Dr. Krupa Rasane 56
Dr. Krupa Rasane 57
Dr. Krupa Rasane 58
Dr. Krupa Rasane 59