Upload
others
View
10
Download
0
Embed Size (px)
Citation preview
Raj JainThe Ohio State University
1
What is a Medium?Storage Dept
Magnetic tape
Floppy disks
Optical disks
Raj JainProfessor of Computer and Information Sciences
The Ohio State UniversityColumbus, OH 43210
[email protected] http://www.cis.ohio-state.edu/~jain/
Raj JainThe Ohio State University
2
Local Multimedia
Video Fundamentals
Compression methods
Compression Standards: JPEG, MPEG,...
Overview
Raj JainThe Ohio State University
3
Multimedia: Key IngredientsTwo or more media: Text, Image, Audio, VideoSynchronization among the mediaHuman users: senders, receivers or bothInteraction: Search, fast forward, conversation
Raj JainThe Ohio State University
4
Multimedia ApplicationsDistance learningHealthcare: Telemedicine, Remote DiagnosisRemote access to health databaseTelepresence: Virtual proximity. Can control remotecamera.
Real estate purchasers can drive down the virtual cityVideo Cruising: Virtually go through coworker's offices,coffee roomsCollaboration: Shared-screen systems
Raj JainThe Ohio State University
5
Multimedia Market
Video Games
Movie Theatres
Arcades
CDs/Tapes/Recs
Information Srvs
Home Video
Cabel TV
TV/Radio ADs
Catalog Shopping
Publishing
Computers
Telecom
667101212
2635
5289
151156
0 50 100 150 200
Video Games
Movie Theatres
Arcades
CDs/Tapes/Recs
Information Srvs
Home Video
Cabel TV
TV/Radio ADs
Catalog Shopping
Publishing
Computers
Telecom
$ Billions
Raj JainThe Ohio State University
6
Television Formats
US/Japan: 525/60 NTSC(National Television Standards Committee )
Europe: 625/50 PAL (Phase Alternating Line ),French/E. Europe: 625/50 SECAM (Seq. Color and Memory )525 lines/frame (481 contain video, 22/vertical interval/fieldcan be used for teletext, closed caption, timestamp, etc.)60 Fields/sec. Fields = Odd lines, even lines (Interlaced)30 frames/sec
Frames
Lines
Raj JainThe Ohio State University
7
HDTV
CIF/SIF
Video Formats
In 525 lpf countries:Video Conferencing ⇒ Common Intermediate Format (CIF) = 352×240Source Input Format (SIF) by MPEG = CIF by CCITTVideo Phone ⇒ Quarter CIF = 176 ppl × 120 lpfBroadcast, Cable, VCR ⇒ CCIR 601 = 704×480HDTV = 1280×720 (Progressive=NI) or 1440×960 (Interlace)24 bits/pixel ⇒ 704×480×30×24 = 242 Mbps, 109 GB/h
QCIF
Raj JainThe Ohio State University
8
Video Compression Considerations
High compression100-200 normal, 2500 possible with fractal methods
Decoding must be simple ⇒ AsymmetricH.261, JPEG, AVI, QuickTime are SymmetricDVI, MPEG are asymmetric
Allow real time encoding/decodingImplementable in software, if possibleAllow random-access, fast forward/reverseTransmission Error-tolerantIntegratable with text, graphic files for storage/mailingScalable: Allow a range of video quality
Raj JainThe Ohio State University
9
Compression StandardsSymmetric Asymmetric
Hardwarebased
JPEGH.261
DVI/IntelMPEGCD-I
Softwarebased
AVIQuickTimeUltimedia
Raj JainThe Ohio State University
10
Video Compression TechniquesReducing the frame rate, lines/frame, pixels/line, bits/pixelUsed for teleconferencing. Not acceptable for entertainmentRedundancies: Spatial, Spectral, TemporalNo loss entropy coding:
Run-length coding: 000011111111..100000=0413505
Huffman coding: Frequent characters with fewer bitsDiscrete Cosine Transform
Only low frequency components are quantizedDifferential Pulse Code Modulation (DPCM)
Sections with large interframe differences are quantizedMotion Compensation (Inter-frame)
Differences from predicted motion are quantized
Raj JainThe Ohio State University
11
Pulse Code Modulation (PCM)Nyquist Sampling Theorem: Sample analog signal at twicethe signal frequency
5.44.1
2.95.34.4
34
63.4
101 100 011101100011 011100110
Quantization Step = Unit of quantizationQuantization error= Error due truncation or roundingAcceptable value depends upon noise and human perception
Raj JainThe Ohio State University
12
Discrete Cosine TransformTime domain to frequency domain transform in twodimensionsWith infinite precision DCT is losslessDC coefficient coded differentially from previous blockOther coefficients entropy encoded using zig-zag sequenceQuantization table specifies step size for each element ⇒ Loss
x
yF(u,v)
Raj JainThe Ohio State University
13
Discrete Cosine TransformTake 8×8 block:Compute 64 coefficientsFor u=0,1,...,7 and v=0,1,...,7F(u,v)=0.25 C(u) C(v) { Σ7
x=0 Σ7y=0 f(x,y) cos[(2x+1)u π/16]
cos[(2y+1)v π/16]}f(u,v)=0.25 {Σ7
x=0 Σ7y=0 C(u) C(v) F(x,y) cos[(2x+1)u π /16]
cos[(2y+1)v π /16]}Where, C(i)=1/sqrt(2) for i=0
=1 otherwiseF(0,0) is the DC, others are ACF(u,v) is small for high values of u,v
Raj JainThe Ohio State University
14
JPEG
Joint Picture Experts Group.Joint ⇒ ISO+CCITTDCT + Quantization + EntropyEntropy = Huffman or ArithmeticS/W coding possible: 10 s for 640X480X30 image on 68040
MovingAreaSelect
DCTHuffman+ Run-length
Quanti-zation
Zig-Zagscan
8 × 8 64 coefficients
Raj JainThe Ohio State University
15
JPEG for ColorColor Images = Multiple componentsThe components may be interleavedEach component is divided into regionsEach region consists of several 8X8 blocksAll blocks of one region of one component followed by oneregion of second component and so onCODECs have up to 4 quantization tables and 4 entropytables active
Raj JainThe Ohio State University
16
JPEG for VideoMany vendors use JPEG for videoAlthough designed only for imagesNo interframe coding ⇒ Fast random access221.184 Mbps for 640X480X30X241:50 compression ⇒ 4.4 MbpsQuarter window ⇒ 1 Mbps
Raj JainThe Ohio State University
17
Motion Prediction
MovingAreaSelect
DCTHuffman+ Run-length
Quanti-zation
MotionPredic-
tion
InverseDCT
InverseQuanti-zation
FrameMemory
Zig-Zagscan
8 × 8 64 coefficients
Raj JainThe Ohio State University
18
MPEG-1
Inter-frame CodingI = Intraframe coded ⇒ Allows random accessP = Predicted from previous P or IB = Bidrectional predictionBandwidth Allocation: I:P:B::5:3:1Uses Motion prediction, DCT coding, quantization, entropy
I B B P B B P B B I P
Raj JainThe Ohio State University
19
MPEG-1 (Continued)ISO/IEC JTC1/SC29/WG11 CD 11172Motion Picture Experts GroupVCR quality video for digital storage 1.5-2 MbpsAsymmetric: coding more complex than decodingCompression ratios of 200 ⇒ VCR quality at 1.5-2 MbpsRatio of 50 ⇒ Broadcast quality at 6 MbpsSpecifies rules for multiplexing audio/video streams32 kbps to 384 kbps mono/stereo audioSynchronization using 33 bit timestamps 90 kHz clock
Raj JainThe Ohio State University
20
MPEG-2MPEG Phase 2: Broadcast quality or better15 Mbps for NTSC, 60 Mbps for HDTV, 4-15 Mbps for VCRCompability: Backward/forward, Different picture formatsSpecifies 3 profiles: Simple, Main, NextSimple does not use bidirectional frames
(No storage ⇒ Low cost, Low delay)Main profile designed for most common applicationsNext profile supports hierarchical codingEach profile in 4 levels
High Level Type 1: HDTV, 1152 lines/frame,1920 pixels/line, 60 fps ⇒ 62.7 Mpps = 60 MbpsHigh Level Type 2: HDTV, 1152 lpf, 1440 ppl, 60 fps⇒ 47 Mpps = 60 Mbps
Raj JainThe Ohio State University
21
MPEG-2 (Continued)Main Level: Full size (576 lpf, 720 ppl), 30 fps⇒ 10.4 Mpps = 15 MbpsLow Level: 1/4 Size (288 lpf, 352 ppl), 30 fps⇒ 2.5 Mpps = 4 Mbps
Video industry will use Main profile/main levelCable industry may use Simple profile/main levelUS HDTV will use Main profile/High levelEuropean HDTV will use Next profile/High levelMPEG-2 allows field or frame coding,MPEG-1 allows only frame codingMPEG-2 allows 3 chrominance formats: 4:4:4, 4:2:2, 4:2:0MPEG-1 allows only 4:2:0
Raj JainThe Ohio State University
22
ITU-T H.261 StandardStarted in 1984 for m×384 kbpsLater p×64 kbps p = 1, 2,...,30VCR quality videoResynchronization at receiver ⇒ Allows transmission overindependent parallel channelsDCT + Quantization + Motion-predicted compressionColor represented by Y, Cb, CrY = Luminance, Cb = Chrominance = B-Y,Cr = Chrominance = R-YSampling rates 4:2:2
Raj JainThe Ohio State University
23
ITU-T H.261 StandardCommon Intermediate Format (CIF) for Video conferencingQuarter CIF (QCIF) for desktop telephony
QCIF CIF
Luminance (y) 144 lpf, 180 ppl 288 lpf, 360 ppl
Chrominance (Cr,Cb) 72 lpf, 90 ppl 144 lpf, 180 ppl
Uncompressed rate 9.115 Mbps 29.97 MbpsAllows dropping frames ⇒ reducing frame rateAt 10 fps, QCIF requires compression ratio of 50 for 64 kbpsp = 1 or 2 ⇒ Face only (Video Phone)p = 6 for teleconferencing
Raj JainThe Ohio State University
24
Subband Hierarchical Coding
Hierarchical codingImage signal is split into frequency bandsEach band is processed differentlyExample: 4 bands LL, LH, HL, HHLL consists of low-frequency band in the horizontal scanand low-frequency in the vertical scanLL contains most useful information, coded using DCTOther bands coded using simple quantizers with a dead zoneDead zone results in long strings of zeros
HH
LL
HH
HHHH
HL HL
LH
LHx
y
x
y
Raj JainThe Ohio State University
25
Summary
Video formats: Lines, pixelsCompression techniques: Huffman, run-length, DCT, MotionpredictionCompression Standards: JPEG, MPEG, H.261Hierarchical coding
Raj JainThe Ohio State University
26
References: BooksD. Minoli and R. Keinath, "Distributed Multimedia ThroughBroadband Communications Services," Artech House,1993, 312 pp.B. Lee, M. Kang, and J. Lee, "BroadbandTelecommunications Technology," Artech House, Boston,1993N. Ohta, "Packet Video," Artech House, Boston, 1994T. Trowt-Bayard, “Video Conferencing: The WholePicture,” Flatiron Publishing, Inc., New York, 1994J. Keyes, Ed., “The McGraw-Hill Multimedia Handbook,”McGraw-Hill, New York, 1994, Chapter 32.J. Ozer, “Video Compression for Multimedia,” AcademicPress, Cambridge, MA, 1995.
Raj JainThe Ohio State University
27
References: PapersIEEE Spectrum, Special issue on Digital Television, April1995. [History and explanation of HDTV]Communications of ACM, Vol. 34, No. 4, Apr 1991.B. Furht, “Multimedia Systems: An Overview,” IEEEMultimedia, Spring 1994, pp. 47-59. [JPEG, MPEG, pX64,Networking]T. Chiang and D. Anastassiou, “Hierarchical Coding ofDigital Television,” IEEE Communications Magazine, May1994, pp. 38-45. [MPEG-2]E. Fox, “Advances in Interactive Digital MultimediaSystems,” IEEE Computer, October 1991, pp. 9-21.
Raj JainThe Ohio State University
28
References: Standards[JPEG] JTC1/SC2/WG10, “Digital Compression andCoding of Continuous-Tone Still Images”[MPEG-1] JTC1/SC2/WG11, “Coding of Moving Picturesand Associated Audio for Digital Storage Media up to about1.5 Mbit/s,” ISO/IEC DIS 11172, 1992.[MPEG-2] “Information Technology- Generic Coding ofMoving Pictures and Associated Audio RecommendationsH.262,” ISO/IEC 13818-2 CD, Nov 1993, Seol.[H.261] CCITT SG XV, “Video coder/decoder foraudiovisual services at p × 64 kbps”
Raj JainThe Ohio State University
29
References: OnlineJPEG image compression: Frequently asked questions,http://www.cis.ohio-state.edu/hypertext/faq/usenet/jpeg-faq/faq.htmlCompresion: Frequently asked questions, ftp rtfm.mit.edu,files:/pub/usenet/news.answers/compression-faq/part[1-3]
Distributed Multimedia Survey: Standards,http://cuiwww.unige.ch/OSG/MultimediaInfo/mmsurvey/standards.html#G.711MPEG: Frequently asked questions,http://www.crs4.it/HTML/LUIGI/MPEG/mpegfaq.html
Raj JainThe Ohio State University
30
AcronymsCD-I Compact Disc InteractiveDCT Discrete Cosine TransformDVI Digital Video InteractiveHDTV High Definition TelevisionIVD Integrated Voice DataJBIG Joint Bilevel Image Experts GroupJPEG Joint Photographic Experts GroupMHEG Multimedia and Hypermedia Experts GroupMIDI Musical Instrument Digital InterfaceMPEG Motion Picture Experts GroupNTSC National Television Standards CommitteePAL Phase Alternating LineSECAM Sequential Color And MemorySMPTE Society for Motion Picture and Television Engineers