Page 44 - Proceedings of the 2018 ITU Kaleidoscope

2018 ITU Kaleidoscope Academic Conference

Display Resolution (disRes): It represents the video 100 − max min 100 − Qmax + Dq − Dt2 − Dt3, 100 , 0 ,
display resolution in number of pixels. , framerate < 24
Coding Resolution (codRes): It is the video encoding

resolution in pixels. 100 − max min 100 − Qmax + Dq, 100 , 0

, framerate ≥ 24
3.4.1 Quantization degradation (19)
The estimated mean opinion score is calculated as:

The Quantization Degradation is defined as follows: MOS = MOSq, if Du = 0 and Dt = 0 (20)

= 100 − ( MOSq ) (5) MOSfromR Q , otherwise

where is described as in [ITU-T G.107], and where MOS and Q are the estimated video encoding

=max ( min , 100 , 0 ) (6) qualities on two different scales: MOS∈ [1,5] and Q ∈

and whereMOSq = q + q . exp ( q . quant ) (7) [0,100].

and MOSq = max ( min MOSq , 5 , 1) (8) The Pv module that estimates the video quality with ITU-T
with q = 4.66, q = -0.07 and q = 0.06 P.1203 model is shown in Figure 1. It combines

quantization, frame rate, display and coding resolution to
3.4.2 Upscaling degradation
estimate the MOS.
The Upscaling Degradation is defined as follows: quant

= . log10 ( . ( scaleFactor – 1) + 1 ) ) (9) Quantization Dq

with = max ( min ( , 100 ), 0 ) (10) module

scaleFactor = max , 1 (11)

where u1 = 72.61 and u2 = 0.32. MOS

fps Dt ion module
3.4.3 Temporal degradation Temporal
module
The Temporal Degradation is defined as follows: Integrat

D −D −D , framerate < 24 disRes

D = 0 , framerate ≥ 24 (12) Upscaling

The pure temporal degradation is given by codRes module Du
∙ ( ∙ )
D = (13) Figure 1 - Pv module in the ITU-T P.1203 model

Compensation variable1, relative to coding impact is 4. PROPOSED ALGORITHM
defined as
D = ∙ ( ∙ ) (14) 4.1 DOUBLE SARSA BASED ADAPTATION

ALGORITHM

Compensation variable2, relative to spatial scaling impact
is given by The algorithm used at the client side for quality adaptation
is described as follows:
∙ ( ∙ )
D = (15)

1. Initialize the number of packets N, learning rate α,
with
discount factor γ, Q-matrices Q and Q .
A
B
D = max ( min ( Dt, 100 ), 0 ) (16) 2. Identify the current state s based on throughput Th

where t1 = 30.98, t2 = 1.29 and t3 = 64.65. value using the mapping between throughput and
states.
3.4.4 Integration 3. Read the resolution codRes, frames per second
framerate at which the video is encoded from the
The Degradation (D) due to the quantization ( ), codec header and determine the current action a

upscaling ( ) and temporal ( ) factors are computed as: using the mapping between throughput and states.

D = max ( min ) + + , 100 , 0) (17) 4. Perform calculations to obtain the quant parameter

The maximum quality in relation to the pure upscaling and 5. (using Mode 2 calculation of ITU-T P.1203).
Calculate the quantization degradation Dq,
temporal degradations is defined as follows: temporal degradation Dt, upscaling degradation Du,

Qmax = 100 − − , framerate < 24 (18) degradation D and the quality Q.
100 − , framerate ≥ 24

The quality (Q) is formulated as: 6. Assign the final Mean Opinion Score MOS
obtained as the reward r gained for the current
Q = 100 – D =
state s and current action a.
– 28 –