Multimedia Communications Spring 2006-07 Video Quality & - PowerPoint PPT Presentation

CS 584 / CMPE 584 Multimedia Communications Spring 2006-07 Video Quality & Traffic Characteristics Shahab Baqai LUMS

Video Compression & Encoding Control Schemes � Quality degradation resulting from – Digitization – Quantization � For fixed given values for the encoder parameters both quality and bit rate vary with scene content � To achieve certain quality, bit rate, and/or delay objectives over time, encoder parameters must be dynamically adjusted � Video Encoding Control Schemes – Constant bit rate – Open-loop variable bit rate – Constant quality variable bit rate 2

ITS Video Quality Measure – ˆ s � Quantitative video quality measure developed by ITS Institute for Telecommunication Sciences � Measures spatial & temporal information loss � 94% correlation with subjective evaluations � Validated using 10 sec sequences, wide range of impairments � Quality degradation measured on a scale of 1 to 5 3

Video Quality Measure – ˆ s = − − ˆ s 4.77 m 0.272 m 0.356 m 1 2 3 where m :spatial distortion measure 1 m : temporal distortion measure, detects lost motion 2 m : temporal distortion measure, detects added (false) motion 3 4

Video Quality Measure – SNR N ∑ p ( ) 2 o n i = = i 1 SNR 10log 10 N p ∑ ( ) ( ) 2 ⎡ − ⎤ o n d n ⎣ ⎦ i i = i 1 where ( ) th th o n : Luminance value for the i pixel of the n orignal frame i ( ) th th : Luminance value for the i pixel of the n encoded/decoded frame o n i : Number of pixels in a frame N p 5

6 Tools & Experimental Setup

Analysis of Video Traffic & Quality Characteristics 7

8 Video Sequences Used

Description of Video Sequences Used � Video Conferencing – Head shoulders, no scene change, little motion � Terminator 2 – Morphing, slow movement, sharp edges, several scene changes � Raiders – Some very fast moving parts (panning), many scene changes � Star Trek – Mixture of fast & slow movement, some irregular movement due to camera shaking, 29 scene changes � Commercials – Mixture of real & synthetic images, second half contains very fast movement and sharp edges 9

Constant bit rate Video Encoder Control 10

11 Quantization

Quantization Matrix for Intra-Coded Macro Block 12

Star Trek, V = 384 kbps, B = 19.2 kbits 13 CBR Quality Characterization

Star Trek, V = 384 kbps, B = 384 kbits 14 CBR Quality Characterization

Star Trek, V = 384 kbps, B = 1920 kbits 15 CBR Quality Characterization

Star Trek, V = 1536 kbps, B = 19.2 kbits 16 CBR Quality Characterization

Star Trek, V = 1536 kbps, B = 384 kbits 17 CBR Quality Characterization

Star Trek, V = 1536 kbps, B = 1920 kbits 18 CBR Quality Characterization

CBR Quality Characterization Video Conference, V = 384 kbps, B = 19.2 kbits 19

CBR Quality Characterization Video Conference, V = 384 kbps, B = 384 kbits 20

CBR Quality Characterization Video Conference, V = 384 kbps, B = 1920 kbits 21

CBR Quality Characterization Video Conference, V = 1536 kbps, B = 19.2 kbits 22

CBR Quality Characterization Video Conference, V = 1536 kbps, B = 384 kbits 23

CBR Quality Characterization Video Conference, V = 1536 kbps, B = 1920 kbits 24

CBR Quality Characterization Commercials, V = 384 kbps, B = 19.2 kbits 25

Commercials, V = 384 kbps, B = 384 kbits 26 CBR Quality Characterization

CBR Quality Characterization Commercials, V = 384 kbps, B = 1920 kbits 27

CBR Quality Characterization Commercials, V = 1536 kbps, B = 19.2 kbits 28

CBR Quality Characterization Commercials, V = 1536 kbps, B = 384 kbits 29

CBR Quality Characterization Commercials, V = 1536 kbps, B = 1920 kbits 30

31 Quality vs. B for Star Trek

32 Quality vs. V for Star Trek

Quality vs. V for All Sequences 33

Equal Contours, Star Trek 34 min ˆ s

Equal Contours, All Sequences 35 min ˆ s

CBR Delay Characterization for Transmission over a circuit � Delay for a Video transmission over a circuit 36

V = 1536 kbps C = V 37 Delay Statistics vs. B V = 384 kbps

C = V 38 Delay Statistics vs. V

C = V 39 Maximum Delay Contours

Meeting Delay & Quality Objectives (1) 40

Meeting Delay & Quality Objectives (2) 41

42 Transmission at C > V (1)

43 Transmission at C > V (2)

CBR: Meeting Quality & Delay Constraints by Choosing V < C � Given D – C = 1024 kbps – D max = 30 ms, s min = 4.2 � If V is chosen to be 1024 kbps, then delay & quality objectives cannot be met eg for – B = 30 kbits, D max = 30 ms s min = 3.5 � Choosing In Particular when Dmax < 40 – 50 ms, V = 768 kbps & B = 384 kbits it is useful to choose V to be somewhat we achieve smaller than C D max = 25 ms s min = 4.2 44

Number of Bits per Frame for Star Trek 45 CBR Traffic Characterization

CBR Traffic Characterization Number of Bits per Frame for Video Conferencing 46

Frame Size Histograms for Star Trek 47 CBR Traffic Characterization

CBR Traffic Characterization Frame Size Histograms for Video Conferencing 48

Frame Size Histograms for Commercials 49 CBR Traffic Characterization

CBR Traffic Characterization Bits per Frame Statistics V = 1536 Kbps, B = 384 kbits (Average number of bits per frame = 51200) 50

Multiplexing CBR Streams over a Circuit 51

Video Conference: Multiplexing Performance 52

Commercials: Multiplexing Performance 53

Frame Size 54 CBR: MPEG

Open-loop Variable bit rate Video Encoder Control 55

56 Quality vs. Time for Star Trek

57 0 q Minimum Quality vs.

58 Number of Bits per Frame

59 0 q Average Frame Size vs.

60 0 q Maximum Frame Size vs.

61 Frame Size Histograms

Frame Size Autocorrelation Functions 62

OL-VBR MPEG Encoding: Frame Size vs. Time 63

Frame Size Histogram: MPEG vs. H.261 64

Frame Size Histograms for I, P, & B Frames 65

Constant Quality Variable bit rate Video Encoder Control 66

Video Conferencing: Constant Quality Achieved 67

68 CQ-VBR: Quality Statistics

CQ-VBR: Number of Bits per Frame 69

70 Frame Size Histograms

71 Frame Size Statistics

Frame Size vs. Target Quality 72

CQ-VBR: Delay when Video is Transmitted over a Circuit 73

74 Maximum Delay vs. C

Comparison: CQ-VBR Achieves Lower Average Bit Rate 75

Comparison: CQ-VBR Achieves Lower Average Bit Rate 76

Joint Peak rate & Quality Controlled VBR 77

JPQC-VBR: Peaks Eliminated 78

JPQC-VBR: Comparison of Quality Statistics with CQ-VBR 79

JPQC-VBR: Comparison of Frame Size Statistics with CQ-VBR 80