Video Transmission

Jerkiness Measure
Deena Abdelsamad

This thesis is presented as part of Degree of

Bachelor of Science in Electrical Engineering

Blekinge Institute of Technology

June 2012

Blekinge Institute of Technology

School of Computing
Supervisor: Raja M. Khurram Shahzad
Examiner: Benny Lovstrom

1
Abstract

Digital video transmission is widely used nowadays in multimedia. Frame drop-

ping, freeze and reduced number of frames in the transmitted video are common
symptoms of bad transmission quality. In order to assess the quality of trans-
mission, a criterion is introduced in a model for a no reference video jerkiness
measure [3]. This model is dierent from the former models presented as it
depends on viewing conditions and video resolutions, so it is applicable for any
frame size from QCIF to HD. The model uses simple mathematical equations
of jerkiness and can be used for any video sequence [3]. A model of reduced
reference method (Qtransmission) which depends on a pre-measured Jerkiness
is introduced as a suggestion of future work.
The algorithm of video jerkiness measure model [3] is used and a MATLAB
program is implemented to measure the jerkiness of the transmitted video. The
program is used to calculate the jerkiness of QCIF video resolution in both (avi)
and (yuv) formats. Twelve test cases are used to test the implemented program;
the rst six test cases are in (avi) format while the remainder cases are in (yuv)
format. The rst nine cases are error free with no observed frame freeze or
drop. Intentional error (freeze) is added in the last three test cases, frame freeze
is forced to occur in almost a third of the video frames and the jerkiness is
calculated accordingly. The implemented program is used to calculate the video
jerkiness, conclusion and results are presented and discussed in chapter 4 of this
thesis.

2
Acknowledgments

Completion of work gives a nice feeling of happiness and success. The accom-
plishment of this thesis gives me the condence to move ahead and teaches me
a new way of knowledge and learning. I would like to convey my gratefulness,
praise and thanks to ALLAH (swt) the most gracious and the most merciful,
who blessed me with his grace and helped me to write this thesis successfully.
Thanks to Anders Hultgren, program manager and Raja M. Khurram Shahzad
my supervisor who provided me with valuable comments and suggestions for
improvement. Thanks are due to all the BTH sta members who are always
friendly and supportive. Last but not least, my sincere gratitude goes to my
parents and my husband for their continuous mental, emotional and nancial
support. I feel lucky and proud to be part of my family.

3
Contents

1 Introduction 8
1.1 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Background and related work 10

2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.1 The History of the Motion Picture . . . . . . . . . . . . . 10
2.1.2 Number of frames per second . . . . . . . . . . . . . . . . 11
2.1.3 Interlacing . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.4 RGB color model . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.5 YUV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Related work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3 Design and Implementation 13

3.1 Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3.1 Reading code . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3.2 Sshape Function . . . . . . . . . . . . . . . . . . . . . . . 18
3.3.3 YUV to MOV Conversion . . . . . . . . . . . . . . . . . . 19
3.3.4 Adding Freeze . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3.5 MOV to YUV Conversion . . . . . . . . . . . . . . . . . . 20
3.4 Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

4 Results, Conclusions and Future work 21

4.1 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.3 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

4
List of Figures

3.3.1 Read Video . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.3.2 Resolution Parameters . . . . . . . . . . . . . . . . . . . . . . . . 15
3.3.3 Resolution Settings . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3.4 mu Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3.5 Motion Intensity Calculation . . . . . . . . . . . . . . . . . . . . 17
3.3.6 Calculating Jerkiness . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.3.7 Sshape function . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3.8 YUV to MOV Conversion . . . . . . . . . . . . . . . . . . . . . . 19
3.3.9 Adding Freeze . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3.10 MOV to YUV Conversion . . . . . . . . . . . . . . . . . . . . . . 20

4.1.1 Test Case 1 Results . . . . . . . . . . . . . . . . . . . . . . . . . . 23

4.1.2 Test Case 2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.1.3 Test Case 3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.1.4 Test Case 4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1.5 Test Case 5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.1.6 Test Case 6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.1.7 Test Case 7 Results . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.1.8 Test Case 8 Results . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.1.9 Test Case 9 Results . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.1.10 Test Case 7 with freeze Results . . . . . . . . . . . . . . . . . . . 32
4.1.11 Test Case 8 with freeze Results . . . . . . . . . . . . . . . . . . . 33
4.1.12 Test Case 9 with freeze Results . . . . . . . . . . . . . . . . . . . 34

5
List of Tables

4.1 Summary of results . . . . . . . . . . . . . . . . . . . . . . . . . 35

6
List of abbreviations

1. HD - High-denition

2. SD - Standard-denition

3. CIF - Common intermediate format

4. QCIF - Quarter common intermediate format

5. AVI - Audio Video Interleave

6. CRT - Cathode ray tube

7. VCR - Video Cassette Recording

8. LCD - Liquid-crystal display

9. DLP - Digital Light Processing

10. VQEG - Video Quality Experts Group

11. DVD - Digital Versatile/Video Disc

12. MPEG - Moving Picture Experts Group

13. VHS - Video Home System

14. PAL - Phase Alternating Line

15. NTSC - National Television System Committee

16. 3D - Three-dimensional

17. PDP - Plasma display panel

18. RGB - Red, Green, Blue

19. SECAM - Sequentiel couleur a memoire ( French for Sequential Color with
Memory )

7
Chapter 1

Introduction

Digital videos transmitted over communication networks are easily subjected

to frame dropping, freezing and reducing. For an end user these impairments
cause a non-smooth presentation of the video that is called Jerkiness. A general
denition of frame freezing is given below.
Frame freezing can be caused by low-bit rate encoding when there is an
overow in the encode buer. It can also be caused by the transmission of the
video stream over an error prone channel [4].
Frame skipping aects image rate freezing and jerky motion especially at low
bitrates. This also aects the video quality at the nal receptor causing image
sharpness loss such as blur and impairments such as freezing and jerkiness. If
the nal receptor is human, it is important to have a no-reference measure
for the received video quality to detect Jerkiness from the output of a video
player. This measure depends only on information taken from the output video
sequence where no information about the video before encoding or transmission
is required [9].
In a study, a model for measuring jerkiness is proposed [3]. Results indicated
that the proposed model is unique and better from existing models of that time
[3]. First, it is applicable for any frame size from quarter common intermediate
format (QCIF) to high denition (HD), that depends on the viewing conditions
and video resolution. Second, the model can be customized and used for any
video sequence. Third, simple mathematical form is used to measure the Jerki-
ness. Model of calculating Jerkiness is used to calculate the variables value, and
then the variables are dened in the implemented program and substituted in
the following formula [3].

1X
J (v) = 4ti (4ti ) (mi+1 (v)) (1.0.1)
T

8
1.1 Scope

The scope of this thesis is to measure jerkiness of an output transmitted video

by implementing an algorithm of a no-reference measure for the received video
quality based on a no-reference model which depends only on information taken
from the output video sequence [3]. A test will be run rst on the implemented
program to check it's eciency to determine frame number and calculate motion
intensity, later on, some video samples are tested with and without freeze to
determine the freeze eect on jerkiness value.

1.2 Outline

Chapter 2 focuses on the basic concepts and background for image and video
history. Chapter 3 describes the design and implementation of the program,
it presents a description of MATLAB commands used and tested. Finally in
Chapter 4, results are presented and their explanation is given. Chapter 4 also
concludes the work and provides directions for future work.

9
Chapter 2

Background and related work

2.1 Background

2.1.1 The History of the Motion Picture

Wheel of life or Zoopraxiscope was the rst machine invented by William

Lincolin in the United States in 1867 that was able to show animated photos
and movies. The motion picture usage started by inventing the motion pictures
cameras by Lumiere 1895 [1]. It included the three following functions, i.e.,
mobile motion-picture camera, projector and movie processing unit. Although
some others had similar inventions at the same point of time but Lumiere and his
brother were considered the rst to present these technologies of photographic,
projected and moving pictures to audience of one person or more [1]. A research
team, led by Charles Ginsburg, at the Ampex Corporation was responsible to
invent a recording magnetic video tape that can capture the live pictures from
the TV camera and converts them into electrical pulses and saves them. The
team succeeded to develop magnetic video tape in 1951 [1]. In 1971 the Sony
company sold the rst video cassette recording (VCR) to consumers [1]. Later,
a massive decrease of video tapes sales took place by invention of digital video
disc (DVD) in 1997 and the Blu-ray in 2006 and video tapes became an old
fashion [1].

Vidre is a Latin verb with the meaning See, while the word
Video typically means I See. But practically it refers to store the
moving photos with dierent digital formats like DVD, moving pic-
ture experts group (MPEG), audio video interleave (AVI) or analog
formats like video home system (VHS) and transmits them with dif-
ferent techniques like phase alternating line (PAL) or national tele-
vision system committee (NTSC) [1]. The quality of a transmitted
video depends on dierent parameters such as how the moving pic-
tures were captured and how they were stored. A modern format of
television video became the standard format which can oer higher

10
 quality than the former ones which is digital television (DTV) [6].

2.1.2 Number of frames per second

The frame rate is the number of pictures per unit of time in the video; it diers
according to the type of the camera used to capture pictures. The normal rate
of old cameras can vary from six to eight frames per second, while it reaches up
to 120 frames per second for modern cameras. It is discovered that the frame
rate aects the transferring process of the cinematic motion picture to a video
lm, as the movie recorded at slow frame rate like 24 photo-grams per second
can make the transmission more complicated. Besides, to get the illusion of a
video movie, without giving the feeling to user that he/she is watching moving
photos, a frame rate of at least 15 frames per is suggested [7].

2.1.3 Interlacing

Video has two scan formats, progressive or interlaced, such as sequential color
with memory (SECAM), NTSC and PAL 576i50. Interlaced video is a method
to double the received frame rate displayed with the signal used with analog
television without consuming more bandwidth [12]. Interlacing is indicated in
the video data as i where 576 is the resolution of the vertical line and 50 is the
elds of half-frames per second. The main usage of interlacing is to get the best
video quality if the bandwidth is limited. In each interlaced frame, horizontal
scan lines are numbered respectively and divided into two elds, i.e., upper eld
and lower eld. The upper eld also called odd eld contains lines with odd
numbers, the lower eld that is also called even eld contains the lines with
even numbers. This interlaced stream like DVD or analog can be converted by
a method called deinterlacing to use it with the progressive devices like liquid
crystal display (LCD) and plasma screens. However, this deinterlacing method
is unable to give the same video quality processed by progressive scan. The
progressive system has a dierent technique, it updates all scan lines with every
refresh period, which enhances the resolution and decreases errors like moving
or ashing of the constant pictures [12].

2.1.4 RGB color model

RGB model is an additive color model where R stands for red, G for green and
B for blue, light is added in dierent ways to reproduce a massive range of the
colors. The main purpose of the RGB model is to display, represent and sense
image in the electronic systems, for example computer screens and televisions
[10]. RGB has dierent input methods such as video, television cameras and
image scanners. RGB has dierent output devices as well such as mobile phones,
computer screens, projectors and televisions with dierent systems like cathode
ray tube (CRT), LCD and plasma [6].

11
2.1.5 YUV

YUV is a color space which encodes videos considering human perception. It

also refers to the complete range of colors that can be recorded and displayed
in a digital video [2]. Y is for Luma (brightness) component while U and V are
for chrominance (color) components respectively. YUV color encoding system
is used for PAL and NTSC. The equations to convert RGB to YUV are [2]:
Y = 0.299R + 0.587G + 0.114B
U = 0.147R - 0.289G + 0.436B
V = 0.615R - 0.515G + 0.100B

2.2 Related work

In the last few years, researches have been done on a non-reference video quality
measure. Some of them presented a new metric to evaluate and detect the eect
of image dropping on user quality perception [8]. That measure was based on
a psycho-visual quality function and temporal summation function (temporal
pooling) modeling the assessment mechanism of the human assessors [8]. This
assessment model integrates the abrupt temporal variation that appears at the
end of uidity impairments as a second factor for quality estimation [8]. While
other similar studies were done about the same type of measure, however, the
measure is based on freezes, jerky motions and rate variations of an image [9].
The measure should show a signicant correlation with the observers ratings in
an attempt to reproduce some basic perceptual human visual process involved
within the task of video quality assessment [9].

12
Chapter 3

Design and Implementation

3.1 Requirements

To measure the jerkiness of a video, a program is required that is capable of

reading video frames and converts them to numerical data (Y, Cr, Cb). The
program must also calculate the motion intensity and evaluate the required
parameters to compute the jerkiness.

3.2 Methodology

In this thesis, the aim is to implement a no-reference measure of jerkiness that

depends on information taken from the output video sequence where no infor-
mation about the video before encoding or transmission is required [10].
Borer's model algorithm and formulas are used in calculating the variables
values. Then, they are dened in the developed program and substituted in the
following formula of calculating Jerkiness.

1X
J (v) = 4ti (4ti ) (mi+1 (v)) (3.2.1)
T

Displayed images with certain rate and time stamps form a video sequence
can be denoted by v = (fi , ti ) , i = 1...n [3]. Where fi is frame  i of the video,
ti is starting displaying time up to ti+1 and n is the total number of frames.
In case that the number of frames is already known, it will be mentioned as
time stamps 4ti = ti+1 ti . It is assumed that calculating motion by the
theory of calculating the statistic velocity distribution of all objects moves in
the video sequence is very complicated, so a simpler measure is used for motion
calculation in a video sequence v = (fi , ti ) by the following formula [3]. Where
fi (x) is the pixel value of the Y-component of the frame  i located at the x
location. s
X 2
mi+1 (v) = (fi+1 (x) fi (x)) (3.2.2)

13
 The jerkiness calculation in Borer's model [3] depends on the number of
frames, frame display time 4ti and frame motion intensity. The dependency on
4ti and the motion intensity are expressed by two S-shaped (sigmoid functions)
and . These functions have three parameters represented in position of
x, position of y and slope of the inection point. For calculating the motion
dependent part , these three parameters are kept xed. For the of display
time dependent part they are re-parameterized by single parameter where
qpx /py
a = py /px , b = qpx /py , c = 4q/d and d = 2 (1 py ).

axb

if x px
s (x) = d (3.2.3)
1+exp(c(xpx)) +1d else

As the algorithms are tested on QCIF resolution videos, so for the param-
eters are (px , py , q) = (5, 0.5, 0.25) and for the parameters are (px , py , q) =
(0.12, 0.05, 1.5). The S-shaped function starts at the origin and increases poly-
nomial until it reaches the inection point then it saturates exponential towards
one. The viewing angles and distances are proved to aect the results; tests are
performed with the typical viewing angle and distance [5]. The viewing distance
is equal to three times the height of the picture (3 H) and viewers are seated
directly in line with the center of the video display. The resolution is set as
QCIF and motion intensity is measured on the sub-sampled frames in case of
larger resolutions.

3.3 Implementation

3.3.1 Reading code

The program for calculating video jerkiness consists of two parts. The main
program is ReadVid.m that contains all the functions required to read the
video, slice it into frames, compute motion intensity and other required values
in order to compute the jerkiness. The second part is Sshape.m, it calculates
the s-shape value required to compute, .

In gure 3.3.1, clear all command is used to clear the MATLAB screen, and
then close all to close any open windows or gures opened in MATLAB. The
code also performs the following:
Read video specied
slice video into frames
convert RGB matrices of each frame to YCbCr matrices

14
clear all
close all
inle = 'clip6.avi';
readerobj = mmreader(inle)
vidFrames = read(readerobj);
numFrames = size(vidFrames,4);
rgb=zeros(1,1,1);
for k = 1 : numFrames
rgb = vidFrames(:,:,:,k);
% transfearing the rgb data to YCbCr
mov(k).cdata=rgb2ycbcr(rgb);
end

Figure 3.3.1: Read Video

info = mmleinfo(inle);
hight=info.Video.Height;
width=info.Video.Width;
qciag=0;
ciag=0;
hdag=0;
sdag=0;
c=0;
%QCIF 176 144
%CIF/SIF(625) 352 288
% HD width is >= 1080
if ((width==176)&&(hight==144))
qciag=1;
c=1;
end
if ((width==352)&&(hight==288))
ciag=1;
c=1.18;
end

Figure 3.3.2: Resolution Parameters

In gure 3.3.2, mmleinfo MATLAB function is used to get the frame width
and height. Each resolution (QCIF, CIF, HD, SD) will have certain constants in
the future calculations. Four ags (qciag, ciag, hdag, sdag) are initialized
to have the value of zero then according to its values of width and height
the appropriate ag are set. For example if width =176 and height = 144 so
the video is of QCIF resolution. Therefore, the qciag is set to 1 while other
ags remain zero.

15
if ((hight==480)||(hight==576))
sdag=1;
c=1.54;
end
if (width>=1080)
hdag=1;
c=2.54;
end
if ((qciag==0)&&(ciag==0)&&(sdag==0)&&(hdag==0))
disp('the vedio resolution is not recognized');
disp('excution terminated');
break;
end

Figure 3.3.3: Resolution Settings

Figure 3.3.3, continues to set the appropriate frame resolution ag. The last
part of the code is to break the execution of the program if the frame resolution
is not recognized.

% the mu parameters [Px,Py,q]

mupar=[5,0.5,0.25];
% the tau(alfa) =[Px/c, Py, q*c]
taupar=[0.12/c,0.05,1.5*c];

Figure 3.3.4: mu Parameters

In gure 3.3.4, as the code is written to measure the jerkiness of QCIF videos,
so the parameters of , of QCIF videos are used. Note that c=1 for QCIF
resolution.

16
% calculating the motion intensity
% for avi Dt is uniform
dt=info.Duration/numFrames;
for k=1:numFrames-1 %
extracting the R matrix of frames F(i) , F(i+1)
Fiy= double(mov(k).cdata(:,:,1));
Fip1y=double(mov(k+1).cdata(:,:,1));
Ficb= double(mov(k).cdata(:,:,2));
Fip1cb=double(mov(k+1).cdata(:,:,2));
Ficr= double(mov(k).cdata(:,:,3));
Fip1cr=double(mov(k+1).cdata(:,:,3));
irow=size(Fiy,1);
jcol=size(Fiy,2);
dumy=0;
dumcb=0;
dumcr=0;
for i=1:irow
for j=1:jcol
dumy=dumy+(Fip1y(i,j)-Fiy(i,j))^2;
dumcb=dumcb+(Fip1cb(i,j)-Ficb(i,j))^2;
dumcr=dumcr+(Fip1cr(i,j)-Ficr(i,j))^2;
end
end
dumy=double(dumy);
dumcb=double(dumcb);
dumcr=double(dumcr);
my(k)=sqrt(dumy)/width;
mcb(k)=sqrt(dumcb)/width;
mcr(k)=sqrt(dumcr)/width;
% mav=sqrt((dumr+dumg+dumb)/3);
end
gure(1)
plot(my,'r');
hold on
%plot(mcb,'g');
%hold on
%plot(mcr,'b');
%hold on
%plot(mav,'k');
title('motion intesity');
ylabel('Motion intensity');
xlabel('Frame no')

Figure 3.3.5: Motion Intensity Calculation

17
 In gure 3.3.5, the code computes the motion intensity and plots it.

% computing the jerkness

imax=max(size(my));
dum=0.0;
for i=1:imax
mu(i)=Sshape(mupar,my(i));
tau(i)=Sshape(taupar,dt);
dum=dum+dt*tau(i)*mu(i);
end
jerkness=1.0/info.Duration*dum
dt
taupar
gure(2)
plot(mu,'r');
title (' mu values')
ylabel('mu');
xlabel('Frame no')
gure (3)
plot(tau,'b');
ylim([0.8*min(tau),1.2*max(tau)])
title ('Tau values')
ylabel('Tau')
xlabel('Frame no')
disp
('nish')

Figure 3.3.6: Calculating Jerkiness

Figure 3.3.6, shows the part that calculates and using Sshape function
which is presented in the next section. The code uses the already derived and
to compute the jerkiness while MATLAB Figures (2,3) plot and respectively
for each frame in the video.

3.3.2 Sshape Function

The Sshape function in gure 3.3.7, computes and of a frame in the video
sequence. It uses the and parameters as an input plus 4t if is required
or motion intensity if is required. The returned value ( or ) is the out
variable in gure 3.3.7.

18
function out = Sshape(invec,x)
px=invec(1);
py=invec (2);
q=invec(3);
b=q*px/py; d=2*(1-py);
c=4*q/d; a= (py/px)^b;
if x<=px out=a*x^b; else out=d/(1+exp(-c*(x-px) ))+1-d;
end
end

Figure 3.3.7: Sshape function

3.3.3 YUV to MOV Conversion

In gure 3.3.8, values are set and the freezing code is executed in the video, then
the video is converted from YUV to MOV [11]. This part of the code should
perform the following:
Delete any former memory in working space
Close all the opened windows
Read le clip9
start from 2/3
end at 3/3
Sample rate = 420
Convert YUV to MOV
Number of frames = size of matrix

clear all
close all
inle = 'clip9';
clip=[2/3,3/3];
samplerate=420;
mov = yuv2mov([inle,'.yuv'],176,144,num2str(samplerate));
numFrames = size(mov,2);
hight=144;
width=176;

Figure 3.3.8: YUV to MOV Conversion

3.3.4 Adding Freeze

In gure 3.3.9, some freeze (repeated frames) is added to the video to compare
the dierent results when the code runs on same video with and without freeze.
This part of the code should perform the following:
Start freeze from istart
avoid decimals
Start from the rst part or number 1

19
 start from (2/3)
stop at (3/3)
Start the freezing from start to stop
K = number of the current frame used
delete the values of frames from start to stop
repeat frame of start
End

istart=round(1+clip(1)*numFrames);
iend=round(clip(2)*numFrames);
for k=istart:iend
mov(k).cdata=mov(istart).cdata;
end

Figure 3.3.9: Adding Freeze

3.3.5 MOV to YUV Conversion

In gure 3.3.10, this function is used to convert MOV object to YUV le [11].
The code takes the old name of the freeze free le and add to it the word -freez,
then it takes the same sample rate in a string using the MATLAB tool number
to string.

mov2yuv([inle,'-freez','.yuv'],mov,num2str(samplerate));

Figure 3.3.10: MOV to YUV Conversion

3.4 Testing

To test the code, freezing algorithm is used to freeze certain parts in the video. It
is expected that the video with some freezing frames gives lower jerkiness value
than the video without freezing. A simple test is made which freezes the entire
video; it contains one frame but repeated 300 times so that all the frames have
the same numerical data. The frame repetition leads to zero motion intensity
for all frames and zero jerkiness as well. The implemented program gives the
zero jerkiness value as expected.

20
Chapter 4

Results, Conclusions and

Future work

4.1 Results

Twelve short videos are used to test the program that computes the jerkiness.
The rst six videos are of avi format and of QCIF resolution, while the remain-
der movies are of YUV format. It should be noted that in avi video format,
dt is constant, i.e, the time interval for each frame display in the video frame
sequence is not changing and causes a constant value of (t) throughout video
display time. That is shown in (c) of all gures. Figures (4.1.1-4.1.12), show
plots of the motion intensity, (m), and (t). Table (4.1) summarizes test cases
details in addition to the resultant jerkiness values for each test case. First six
(1- 6) videos are tested normally without any errors and three (7- 9) videos are
tested before and after adding freeze to them. Figure 4.1.1, shows the results of
clip1.avi. It shows that this clip has high motion intensity for frames 150 and
above, this leads to a (m) close to 1 for those frames as shown in (b). This
clip is of avi format so 4t is uniform, therefor (t) is constant. Figure 4.1.2,
shows the results of clip2.avi, it shows also that the number of frames is larger
(~1200). This clip is of avi format, 4t is uniform, therefor (t) is constant but
higher than (t) of clip1.avi because of higher number of frames. The jerki-
ness value of clip2.avi is higher than clip1.avi because it has larger number
of frames, a signicant portion of frames has high value of motion intensity and
(m) close to 1. Figure 4.1.3, shows results of clip3.avi, the number of frames
is comparable to clip1.avi but the time step is larger so (t) is larger than
of clip1.avi. The jerkiness of clip3 is 50% larger than of clip1 due to higher
value of (t) mainly. Figure 4.1.4, shows the results of clip4.avi, the rst 60
frames shows almost zero motion intensity and zero (m). Most of the frames
after that shows a high value of motion intensity and a unit (m). The number
of frames is larger than of clip1, (t) is constant and of comparable value to
clip1. The number of frames of almost unit (m) in clip4 is larger than of clip1

21
and all these factors make the jerkiness of clip4 higher than it was in clip1.
Figure 4.1.(5-6), show results of clip5.avi and clip6.avi, there is a high value
of motion intensity for the frames 180~340. For QCIF video, a high motion
intensity means value above 5 and (m) is computed using the following second
equation of the s-shaped function as described in chapter 3. (m) is almost unit
for these frames because motion intensity is higher than 10.

d
(m) = +1d
1 + exp(c (m pm))

Figure 4.1.7, shows the results of clip7.yuv, it also shows that most of the
frames have small value of motion intensity therefore a small value of (t). This
video consists of 300 frames which is larger than the number of frames of clip1.
The resulting jerkiness value of clip7 is smaller than of clip1 due to smaller
motion intensities of the frames of clip7 compared to clip1 although clip7 has
more frames than of clip1. Figure 4.1.8, shows results of clip8.yuv, this video
consists of 150 frames which is half the number of frames of clip7.yuv. Same
value of 4t is observed for all yuv clips used in this study. Figures (4.1.8 b,
4.1.7 b), show that the number of frames with value higher than 5 in clip8 is
larger than clip7 and this is why the jerkiness of clip8.yuv is larger than the
jerkiness of clip7.yuv. Figure 4.1.9, shows the results of clip9.yuv, this video
has 300 frames as clip8.yuv and has the same frame 4t. Comparing Figures
(4.1.8 b, 4.1.9 b), we can see that clip9 has higher average (m) compared to
clip8 and the result of that is a higher jerkiness value of clip9 compared to
clip8. Figures 4.1.(10-11-12), show results of clips7/8/9 with freezing of 1/3 of
its frames. Freezing frames are made by a freeze program that allows dening
the freezing frames interval and replace the numerical data of all the frames in
the freeze interval by the data of the rst frame in that interval. The freezing
procedure causes zero motion intensities of the frames in the freezing interval
and this causes zero value of (m). The jerkiness value of the videos with
intentional frame interval freeze is lower than the jerkiness value of the videos
without freezing as shown in Table (4.1) when comparing case (7 and 10), (8
and 11), and (9 and 12).

22
 motion intesity
18

16

14

12
Motion intensity

10

0
0 20 40 60 80 100 120 140 160 180
Frame no

(a) Motion Intensity

mu values
1

0.9

0.8

0.7

0.6
mu

0.5

0.4

0.3

0.2

0.1

0
0 20 40 60 80 100 120 140 160 180
Frame no

(b)(t)
7 Tau values
x 10

4.6

4.4

4.2
Tau

3.8

3.6

3.4
23
3.2
0 20 40 60 80 100 120 140 160 180
Frame no

(c) (t)

Figure 4.1.1: Test Case 1 Results

 motion intesity
80

70

60

Motion intensity 50

40

30

20

10

0
0 200 400 600 800 1000 1200
Frame no

(a) Motion Intensity

mu values
1

0.9

0.8

0.7

0.6
mu

0.5

0.4

0.3

0.2

0.1

0
0 200 400 600 800 1000 1200
Frame no

(b)(t)
7 Tau values
x 10
4.8

4.6

4.4

4.2
Tau

3.8

3.6

3.4 24

0 200 400 600 800 1000 1200

Frame no

(c) (t)

Figure 4.1.2: Test Case 2 Results

 motion intesity
30

25

20
Motion intensity

15

10

0
0 50 100 150 200
Frame no

(a) Motion Intensity

mu values
1

0.9

0.8

0.7

0.6
mu

0.5

0.4

0.3

0.2

0.1

0
0 50 100 150 200
Frame no

(b)(t)
7 Tau values
x 10

5.2

4.8

4.6
Tau

4.4

4.2

3.8
25
3.6
0 50 100 150 200
Frame no

(c) (t)

Figure 4.1.3: Test Case 3 Results

 motion intesity
35

30

25

Motion intensity
20

15

10

0
0 100 200 300 400 500
Frame no

(a) Motion Intensity

mu values
1

0.9

0.8

0.7

0.6
mu

0.5

0.4

0.3

0.2

0.1

0
0 100 200 300 400 500
Frame no

(b)(t)
7 Tau values
x 10
4.8

4.6

4.4

4.2
Tau

3.8

3.6

3.4

0 100 200 300 400 500

Frame no

(c) (t)

Figure 4.1.4: Test Case 4 Results

26
 motion intesity
50

45

40

35
Motion intensity
30

25

20

15

10

0
0 50 100 150 200 250 300 350 400
Frame no

(a) Motion Intensity

mu values
1

0.9

0.8

0.7

0.6
mu

0.5

0.4

0.3

0.2

0.1

0
0 50 100 150 200 250 300 350 400
Frame no

(b)(t)
7 Tau values
x 10

5.2

4.8

4.6
Tau

4.4

4.2

3.8
27
3.6
0 50 100 150 200 250 300 350 400
Frame no

(c) (t)

Figure 4.1.5: Test Case 5 Results

 motion intesity
70

60

50

Motion intensity
40

30

20

10

0
0 100 200 300 400 500 600 700 800 900
Frame no

(a) Motion Intensity

mu values
1

0.9

0.8

0.7

0.6
mu

0.5

0.4

0.3

0.2

0.1

0
0 100 200 300 400 500 600 700 800 900
Frame no

(b)(t)
7 Tau values
x 10

4.8

4.6

4.4

4.2
Tau

3.8

3.6

3.4 28

0 100 200 300 400 500 600 700 800 900

Frame no

(c) (t)

Figure 4.1.6: Test Case 6 Results

 motion intesity
6

4
Motion intensity

0
0 50 100 150 200 250 300
Frame no

(a) Motion Intensity

mu values
0.7

0.6

0.5

0.4
mu

0.3

0.2

0.1

0
0 50 100 150 200 250 300
Frame no

(b)(t)
7 Tau values
x 10
10.5

10

9.5

9
Tau

8.5

7.5
29

0 50 100 150 200 250 300

Frame no

(c) (t)

Figure 4.1.7: Test Case 7 Results

 motion intesity
9

Motion intensity
6

2
0 50 100 150
Frame no

(a) Motion Intensity

mu values
1

0.9

0.8

0.7

0.6
mu

0.5

0.4

0.3

0.2

0.1

0
0 50 100 150
Frame no

(b)(t)
7 Tau values
x 10
10.5

10

9.5

9
Tau

8.5

7.5
30

0 50 100 150
Frame no

(c) (t)

Figure 4.1.8: Test Case 8 Results

 motion intesity
45

40

35

30
Motion intensity

25

20

15

10

0
0 50 100 150 200 250 300
Frame no

(a) Motion Intensity

mu values
1

0.9

0.8

0.7

0.6
mu

0.5

0.4

0.3

0.2

0.1

0
0 50 100 150 200 250 300
Frame no

(b)(t)
7 Tau values
x 10
10.5

10

9.5

9
Tau

8.5

7.5
31

0 50 100 150 200 250 300

Frame no

(c) (t)

Figure 4.1.9: Test Case 9 Results

 motion intesity
20

18

16

14
Motion intensity
12

10

0
0 50 100 150 200 250 300
Frame no

(a) Motion Intensity

mu values
1

0.9

0.8

0.7

0.6
mu

0.5

0.4

0.3

0.2

0.1

0
0 50 100 150 200 250 300
Frame no

(b)(t)
7 Tau values
x 10
10.5

10

9.5

9
Tau

8.5

7.5
32

0 50 100 150 200 250 300

Frame no

(c) (t)

Figure 4.1.10: Test Case 7 with freeze Results

 motion intesity
50

45

40

35
Motion intensity
30

25

20

15

10

0
0 50 100 150
Frame no

(a) Motion Intensity

mu values
1

0.9

0.8

0.7

0.6
mu

0.5

0.4

0.3

0.2

0.1

0
0 50 100 150
Frame no

(b)(t)
7 Tau values
x 10
10.5

10

9.5

9
Tau

8.5

7.5
33

0 50 100 150
Frame no

(c) (t)

Figure 4.1.11: Test Case 8 with freeze Results

 motion intesity
45

40

35

30
Motion intensity

25

20

15

10

0
0 50 100 150 200 250 300
Frame no

(a) Motion Intensity

mu values
1

0.9

0.8

0.7

0.6
mu

0.5

0.4

0.3

0.2

0.1

0
0 50 100 150 200 250 300
Frame no

(b)(t)
7 Tau values
x 10
10.5

10

9.5

9
Tau

8.5

7.5
34

0 50 100 150 200 250 300

Frame no

(c) (t)

Figure 4.1.12: Test Case 9 with freeze Results

case No of Frames 4t Jerkiness

1 181 0.0400 1.0726 E-7

2 1157 0.0402 2.5963 E-7
3 197 0.0412 1.5205 E-7
4 465 0.0402 2.1209 E-7
5 375 0.0414 2.6523 E-7
6 874 0.0403 1.8876 E-7
7 300 0.05 3.1592 E-8
8 150 0.05 2.0844 E-7
9 300 0.05 5.6480 E-7
7+f 300 0.05 1.8923 E-8
8+f 150 0.05 5.2312 E-8
9+f 300 0.05 3.5826 E-7

Table 4.1: Summary of results

35
4.2 Conclusion

A MATLAB code is implemented to compute video jerkiness to help in video

transmission quality assessment. Table(4.1) shows that the video jerkiness for
videos without frame freeze (cases 7,8,9) is higher than the jerkiness value with
frame freeze (cases 10,11,12) which is an indication of successful implementation
of jerkiness computation procedure presented by Borer [3]. The motion intensity
of the frames in the freezing interval causes zero value of (m) so by substituting
motion intensity values in the following model equation some equations values
are almost equal zero, and after summation the total value are less that the
value of summation equations with more motion intensity values.

1X
J (v) = 4ti (4ti ) (mi+1 (v)) (4.2.1)
T
It is proved also that all the YUV videos used have the same 4t while each
AVI video have its own constant 4t through all the video duration. 4t is shown
to be directly proportional to as calculated in Sshape function as well. For
QCIF video, a high motion intensity means value above 5 and (m) is computed
using the following second equation of the s-shaped function as described in
previous chapters. The (m) is almost unit (1) when motion intensity is higher
than 10.
d
(m) = +1d
1 + exp(c (m pm))

4.3 Future work

This thesis implements a program to measure Jerkiness in video samples of

QCIF resolution. As a future work, this program can be extended to deal with
other formats videos and higher resolutions than QCIF. The implemented pro-
gram can be used for video transmission quality assessment via MATLAB, to
compute the dierent jerkiness values and develop a database to help study-
ing the Jerkiness values of dierent videos with dierent formats, frame rate,
resolutions and 4ts.
It worth to be mentioned, that jerkiness calculation can be used as a quality
measure of transmission. If the jerkiness value before and after transmission is
calculated, i.e., Jbef ore and Jaf ter , then the percentage of quality of transmission
can be calculated as the following:

Jaf ter
Qtransmission = Jbef ore 100 %

Example:
Assume the jerkiness before transmission is substituted by jerkiness value
of a freeze free video (jerkiness value of test case 7 freeze free), while jerkiness
after transmission is substituted by jerkiness value of the same video after adding
freeze (jerkiness value of test case 7 with freeze).

36
 1.8923e008
Qtransmission = 3.1592e008 100 %

Qtransmission =59.9 %

Dierent programming languages can be used as well to develop similar

algorithms to help reaching the same aim which is to deliver the best video
quality to an end user who uses various communications systems.

37
Bibliography

[1] Gina De Angelis. Motion Pictures. The Oliver Press, 2003.

[2] K. Blair Benson. Television Engineering Handbook. McGraw-Hill, 1986.

[3] S. Borer. A model of jerkiness for temporal impairments in video transmis-

sion. pages 218223, 2010.

[4] Mohammed Ghanbari and Quan Huynh-Thu. No-reference temporal qual-

ity metric for video impaired by frame freeze artefacts. 2010.

[5] VQEG HDTV Group. VQEG HDTV Final Report version2.0 video quality
models. volume 2.5, 2008. http://www.its.bldrdoc.gov/vqeg/projects.aspx.

[6] W. Wharton & D. Howorth. Principles of Television Reception. Pitman

Publishing, 1971.

[7] Anil K. Jain. Fundamentals of digital image processing. Prentice Hall, 1989.

[8] Ricardo R. Pastrana-Vidal and Jean-Charles Gicquel. Automatic quality

assessment of video uidity using a no-reference metric. 2007.

[9] Ricardo R. Pastrana-Vidal and Jean-Charles Gicquel. A no-reference video

quality metric based on a human assessment model. 2009.

[10] Charles A. Poynton. Digital Video and HDTV: Algorithms and Interfaces.
Morgan Kaufmann, 2003.

[11] Dima Prfrock. Yuv to matlab movie. MATLAB, 2011.

http://www.mathworks.se/matlabcentral/leexchange/authors/24073.

[12] S. Winkler. Digital Video Quality: Vision Models and Metrics. John Wiley
and Sons, 2005.

38