Technion- Israel Institute of Technology Faculty of Electrical Engineering
CCIT-Computer Network Laboratory
The Influence of Packet Loss On Video Quality
By: Revital and Merav Huber.Supervisor: Dr. Ofer Hadar.
Winter 1998/9
Network Simulation
Analog video Encoder Transmitter
Noisy communication
network
Receiver Decoder Analog video
Original movie Received movie
Target
• Finding the connection between communication network performance and the quality of a compressed video signal.
• Destination: determination of the packet loss model which causes the smallest degradation of the received movie.
Loss Models of The Network
lost. is packet single
passes. packet
]1,0[
PeR
PeR
R
[1,10] burst of size :D
lost, is packet)loss massage(
passes. packet)loss massage(
size). burst average/()loss massage(
raw
R<P
PR
PeP
Assumption
• Bursts of information loss would be less noticed than other error forms.
• reasons:– Error in a number of pixels on consequent
frames, is noticed by viewer more than a frame loss, because of movie speed.
– When bursts occur, following packets are degraded, which belong to the same frame.
Theoretical Background
• MPEG– Motion Picture Experts Group– Video & audio signals synchronization.– Video transfer by frames.– Frame types are determined by encoding:
• Intracoded: encoded by JPEG. • Predictive: use of block- differences between the
previous frame and the present frame.• Bidirectional: use of block- differences between the
encoded frame & the previous & the next frames.
Theoretical Background, MPEG
I frame
Bframe B
frameP
frameB
frameB
frameP
frameB
frameB
frameP
frame
MPEG Movies
Cirinout.mpg
Received.mpg
Cirinout.mpg
Received.mpg
cirinout
received
Theoretical Background- Quality Factors
– Delay – Quantization of MPEG encoding – Lossy network
• causes:– errors due to broadcast bursts.
– Routing errors.
– When buffers are finite and there is overflow, it is possible packets loss will occur.
Performance Idea
• A lossy network simulation, transmitting a video signal.
• The video signal is divided into packets.• The receiver’s information is the arriving packets. • Comparison between the transmitted and received
signals, by PSNR. • Measuring edge sharpness by MTF.• Measuring power spectrum.
Tested Video Signal
Quality Measurements
eM N
sent n m received n m
PSNRP
e
MSE
m
M
n
N
MSE
1
10
2
11
10
2
** [ ( , ) ( , )]
log ( )
(x)
onse(x))(edge_respFFTMTF(f)
xx
j j
j j n
n
S e eX
X e x n e
( ) ( )
( ) [ ]
2
MTF
4. MTF
3. LSF
1. Image
2. Gray Level
1
x x
x x
f f
1
Ideal edge Smeared Edge
False response
Results - Example
Results - PSNR
Pe=0.02Burst IID
Results - MTF
Pe=0.01Burst IID
Results - Power Spectrum
Burst IIDPe=0.02
Conclusions• Burst loss model effects less frames of the movie
than IID.
• Problems:– MATLAB created 2D frames, instead of 3D.– Cut -off video signals during decoding.– Adding lost pictures and erasing created ones distorted
some of the results.
• Though the used quality measurements aren’t ideal, they agree with the theory, and represented well the quality a human eye would notice.
Summary - PSNR
• PSNR isn’t the ideal measurement of the received movie quality. Reasons:– Each frame is analyzed separately, with no
consideration of time sequence. A measurement for video quality should consider time sequence.
– The PSNR result doesn’t determine the picture quality, because it uses a comparison, instead of an absolute value.
Summary - MTF
• The distortion of the signals isn’t linear, thus MTF doesn’t represent only the smearing of the edge response.
• Legal frame: MTF(f=0)=1. Other values represent other forms of degradation.
• Number of legal frames is a very small portion of the total number of frames in the movie.
Summary- Power Spectrum
• Conclusion is achieved by comparing results of several simulations on a vector, with different parameters.
The End