Genista Codec Optimisation_McNally

8/10/2019 Genista Codec Optimisation_McNally

1/16

ITU-T VICA Workshop22-23 July 2005, ITU Headquarter, Geneva

International Telecommunication Union

Perceptual EncoderPerceptual EncoderOptimizationOptimizationDavid McNally

Genista Corp.


2/16

2ITU-T VICA Workshop

22-23 July 2005, ITU Headquarter, Genevadates

ITU-T

CODECs

o Video codecs have reached a veryhigh level of sophistication

o We have two mathematicallyorthogonal mainstream approaches:

1. DCT with motion estimation & F/Bprediction (MPEG family)

2. Wavelet based schemes (JPEG2000)

We observe similar quality as functionof target bit-rate & frame rate:


3/16



ITU-T

CODECs

DCT: Wavelet:

Coding

Transmission


4/164ITU-T VICA Workshop


ITU-T

CODECs

o The currencies for achievable quality

are:

Encoder usage: 1-pass vs. n-pass

Target Bitrate, frame size & frame rate Network behavior (BER, PLR, Latency, Jitter)

BLERvs.

Perceived Quality




ITU-T

CODECs

Encoders have many(!) configurationparameters

! HOW TO CHOOSE OPTIMAL SETTINGS?




ITU-T

CODECs

o CODECs are designed to handleexpected content well:

Sports; Movies; News; Nature;

o They fail for unexpected content:

Tape noise Tape drop-out




ITU-T

CODECs

Encoders are optimized for mainstreamcontent

! HOW TO HANDLE OUTLIERS?


8/168


dates

ITU-T

The Bottom-Line

o The customer does not care whattechnology is used

o The customer has expectations whichdepend on his setting & experience:

IPTV / PPV / ! Broadcast TV

Mobile / Wireless ! Internet streaming

Personalized ! High tolerance

Does the consumer feel he is getting goodvalue for money?!


9/169


dates

ITU-T

The Bottom-Line

o There is an industry which has

developed around codec performance

assessments (e.g. VQEG):

! (Subjective) Test methodologies

! Quality algorithms

! Statistical procedures! Vested interests

S bj i i


10/1610


dates

ITU-T

Subjective testing

0

10

20

30

40

50

60

70

80

90

100

00:00:00:00

00:04:00:12

00:04:50:12

00:05:40:12

00:06:39:24

00:07:29:24

00:08:19:24

00:27:19:12

00:28:09:12

00:28:59:12

00:29:49:12

Time

Scores

32 kbps 128 kb s 384 kb s

17 LCD monitor

viewer

desk

Viewing room (*)

viewing

location 1

curtain

wall background

viewing

location 2

viewing

location 3

Mean Opinion Score

Variance(MOS)

M th d l i


11/16



ITU-T

Methodologies

No Ref erence: Measure subscri ber per ceived qual i t y

Reference Process the content Processed Measure Quality

Reference Process the content Processed

Alignment

Measure Quality

Ful l Ref erence: Measure r elat i ve degr adat i on

Vid Q lit M t i


12/16



ITU-T

Video Quality Metrics

No Reference:

o Blur

o Jerkinesso Blockiness

o Colorfulness

o MOS

Full Reference:

o Blur

o Jerkiness

o Blockiness

o Colorfulnesso Noise

o ANSI metrics

o PSNR

o MOS

NR MOS P di ti P


13/16



ITU-T

NR-MOS: Predictive Power

E ode O ti i tio


14/16



ITU-T

Encoder Optimization

Reference

Parameters

ProcessedENCODER NR

FR

NR

!: Technical optimization

": Service optimization

#: Content optimization

!

"

#

There is Really a Difference!


15/16



ITU-T

There is Really a Difference!

NR Blur

NR Blur

NR Blockiness

WindowsMedia

AppleQuicktime

NR Blkss

Conclusions


16/16



ITU-T

Conclusions

o Good Perceived quality is the key to successfulmedia services

o Encoders are mature and are reaching

asymptotic performance limito Video content is highly variable

o User expectations are highly variable

o Provide cost-performance optimized services -

Need to integrate:

Knowledge of content

Knowledge of user perception & expectation

Knowledge of encoder implementation

Documents

Genista Codec Optimisation_McNally