CC BY
15
\ ИНФОРМАЦИОННЫЕ КАНАЛЫ И СРЕДЫ
UDC 621.39
doi:10.15217/issnl684-8853.2017.2.67
OPTIMIZATION OF ERROR CONCEALMENT BASED ON ANALYSIS OF FADING TYPES
Part 2: Modified and New Models ofVideo Signal Error Concealment. Practical Simulations and their Results
Ofer Hadara, PhD, Associate Professor, hadar@cse.bgu.ac.il Irina Bronfmana, TeachingAssistant, irinamo@post.bgu.ac.il
Nathan Blaunsteina'b, Dr. Sc., Phys.-Math., Professor, nathan.blaunstein@hotmail.com aBen-Gurion Universityofthe Negev, POB 653,1, Ben Gurion St., BeerSheva, 84105, Israel bJerusalem College ofTechnology—Lev Academic Center, 21 Havaad Haleumi, P.O.B. 16031, Jerusalem, 91160, Israel
Purpose: This work is based on the recent research investigations in the combination of two subjects: Fading and Error Concealment. The main aim of the work is to present a more effective method of calculations of fading channel's parameters and to devise methods of achieving of better and more effective performance of Error Concealment, which will lead to higher quality of the video signals after passing through the fading channel. Methods: We explore the influence of fading on a communication channel, by studying the Gaussian, Gaussian, and Ricean distributions. Additionally, we explore existing methods of prediction and of Error Concealment and their influence on the video quality after its exit from the fading channel. Results: It is demonstrated that the Ricean distribution is broader and that it includes the other distributions, Gaussian (ideal channel) and Rayleigh (channel with a strong fading). Therefore, this distribution is used for tests of practical cases occurring in the video channel. On the issue of Error Concealment, a method of Symmetrical CALIC which is an optimization of the CALIC method, was implemented and compared with the original CALIC and with other methods. It has been determined that the proposed optimization yields better results than all the methods used for comparison. In addition, a new method of Error Concealment, named Balanced Percentage Calculation, is proposed. In comparison, its results are two times better on average than the results of Symmetrical CALIC, and are much better than the results of other methods used. Two themes have been combined in such a way that fading influenced the appearance oferrors in the video. Those errors have been replaced by the proposed methods of Error Concealment. All practical tests and comparisons were performed using the MatLab. Practical relevance: The proposed method of calculations of fading channel's parameters allows to perform calculations for all types of channels. This method significantly facilitates work with channels in general and with necessary calculations for channels in particular. The suggested optimization of the existing method of Error Concealment and the new proposed method of Error Concealment allow to receive higher quality video after passing through the fading channel.
Keywords — Ricean Distribution, Ricean Fading, Bit Error Rate, Error Concealment, Level Crossing Rate, Average Fade Duration, CALIC, Symmetrical CALIC, Balanced Percentage Calculation, Weighted Averaging, Intra Prediction using System of Linear Equations.
Modified and New Model of Error Concealment in Channel with Fading
Most existing methods of Error Concealment (EC) use only a part of possible information — one row and one column or two rows and two columns on the same side of the lost block (i. e. a few pixels from two sides are used for recovering one or more of the lost pixels in block, and on the basis of these recovered pixels, the following lost pixels in this block are recovered), a few pixels around the lost pixel located at a certain distance from it (the same pixel row and column is used to recover lost pixels in the same row or the same column, with only changes in distances occurring), etc. [1-10]. We propose using symmetric known information — rows and columns surrounding a lost block. In this case, lost pixels "get" more information and may be recovered more accurately.
The offered method is called Symmetrical CALIC. During the research, another success-
ful method has been found. It is called Balanced Percentage Calculation. An algorithm may be needed in order to optimize it. Both these methods function simultaneously from the four sides of the lost block. Thus, there is simultaneous recover of four pixels, each based on known information or on previously reconstructed pixels.
Symmetrical CALIC. Symmetrical CALIC (SCALIC) method is based on original CALIC method [1], and, as mentioned above, it is performed simultaneously from all four corners (left top down, right top down, right bottom up and left bottom up scanning), as shown in Fig. 1.
In Fig. 1 neighboring pixels that are used in prediction and modeling and the estimated gradients of the image are showed, when P1, P5, P9, P13 are the first (in missed block) predicted pixels
I1, J]' Ip6 I1, J + 3]' IP9[l + 3' 1 + 3]' IP1S + 3' /]• let
us mark them Ipi, Ip&, Ipa, I .
It should be noted that for the convenience of calculation each predictable pixel is marked as Ip[i, j],
nn nn e nnw 1 nn
nw n ne nw n ne
ww w P6 P5 e ee
P1 P2
P3 P4 P8 P7
P15 P16 P12 P11
P13 P14 P10 P9
ww w e ee
sw s se sw s se
ss sse s sw ss
Fig. 1. Symmetrical CALIC scheme
B
50 % T
R
Q
50
P1
P3
P15
P13
P
O
P2
D
E
P4
F
P6
P8
P16
P14
P12
N
P10
M
P5
P7
G
P11
P9
H
J
K
Fig. 2. Balanced Percentage Calculation scheme
when the pixels for its prediction are marked respectively and the real indexes (as above) are placed in i and j. The algorithm for calculating each pixel from missed block is presented in Section "The Simulation Algorithm" and the results of this method are presented in Section "Results of Computation".
Balanced Percentage Calculation. Additional suggested model of EC is Balanced Percentage Calculation. We propose a model based only on two
known pixels — top or bottom (50 %) and side (50 %) as shown on Fig. 2.
In this method, the prediction starts simultaneously with angular pixels and the scanning takes place in the following order: left top down, right top down, right bottom up and left bottom up, as in suggested above SCALIC. The first pixels (angular) are predicted based only on known pixel around lost block, and other pixels are predicted based on known and earlier predicted pixels.
The Simulation Algorithm
H.264 is a modern video format. Videos in this format are used to test the proposed methods of EC and their analysis. Six fragments from various movies in this format are utilized for testing and analysis.
The system shown in Fig. 3 is built in MatLab to test and analyze the proposed methods of BER determining and EC implementation.
First, channel parameters are entered into the system. The following parameters were taken into account to obtain sufficient sample: k = [e-3, 5e-3, e-2, 1], fm = [500, 100, 200] Hz,
ляля 6' 4' 3' 2
[rad], a0 =
n n 6, 3
[rad],
K = [0,1 (close to Rayleigh), 1 (Rice), 10 (close to Gauss)], when the values of bit time and threshold are: Tb = 1,2e-3s, X = 0,001.
Thereafter, for all possible combinations of these parameters, LCR and BER are calculated using the formulas (22), (33) (see Part 1). For convenience, the formulas are presented below again:
^ (x e"(x2 v ^
К (0)
i ^ cosh
Jo
2Xp cos a
К (0)
4psin a +^psin aQ (|psin a)
N т, 1 BER = ^, R = —■ R Tb
d a,
S
I
L
Video in format Channal H.264
Parameters (Compressed Video) Final Video
■ Fig. 3. Simulating system
ИНФОРМАЦИОННЫЕ КАНАЛЫ И СРЕДЫ
All the BERs calculated earlier and the original video were inserted into the channel.
Inside the channel a randomly picked BER value for Ricean fading (for K = 1) is selected. This BER value represents the number of lost bits in each frame. Then, depending on the number of bits per pixel, the number of lost pixels is calculated for each frame. The number of bits per pixel is 24 in all the films used in the simulation.
In H.264 video format, during the passage of the video through the channel, the blocks that are lost consist mostly of 4 x 4 pixels. Therefore, by dividing the previously calculated number of lost pixels by 16 pixels per block, the number of lost blocks in each frame is obtained.
Thereafter, the coordinates of lost blocks are determined randomly. The pixels values in these blocks are nullified and a video with lost blocks, i. e. a corrupted video, ensues. Then, various methods of substitution errors (EC) are used. The outcome of EC constitutes the final, repaired, video.
Error Concealment Symmetric CALIC Algorithm. The pixels used to predict the pixel Ipi [¿, /] are identical to pixels used to predict the pixel Ip[i, j], in the original model:
In = I[i - 1, j], Iw = I[i, j - 1], Ine = I[i - 1, j + 1], Inw = I[i - 1, j - 1], Inn = I[i - 2, j], Iww = I[i, j - 2], Inne = I[i - 2, j + 1].
The gradient of intensity function at the current pixel I is estimated by computing the following quantities:
\IW In Inw\In |;
_\IW ~ Inw | \ln ~ ^яя | ~ Inne \ •
(35)
And the prediction, i. e. concealment, is produced by the following procedure (identical to procedure in the original model):
if - dhi > 80)% sharp horizontal edge
(36)
I Pi I w;
else if {dy^ - dhi <-80j% sharp vertical edge
else {
IPi In;
Л
2 4
if ('dvi - dhi > 32horizontal edge
IPi =
IP1 + Iw
else if - dhi > 8j°% weak horizontal edge
3J„. +
^ =
ip1
else if {^dVi - dh^ <- 32)°%> vertical edge
I n. +
^ =
Pl n
else if - dhi <- 8j% weak vertical edge
^ =
3 Ip, + In
The pixels used for prediction the pixel
TP, MI ■ • • are: In = ip - 1 j], Ie = I[U j + ^
'i=i,i=j+3
Inw = I[i - 1, j - 1], Ine = I[i - 1, j + 1], Inn = I[i - 2, j],
Iee = I[i, j + 2], Innw = I[i - 2, j - 1].
The gradient of intensity function at the current pixel I is estimated by computing the following quantities:
^ftg _ \le ~ I ee\ ^\ln ~ Ine \ \Inw ~ In |'
^Vk \le Ine I In Inn\ Inw In
(37)
And the prediction, i. e. concealment, is produced by the following procedure:
if
- dh >
80)°%> sharp horizontal edge
Ip& Ie;
(38)
else if {dv& - dh& <- 80)%%> sharp vertical edge
else {
т _ Ie In , I"'" I'
1Ръ~ -
n i *-nw *-ne .
4 ;
if - dh& > 32)%%> horizontal edge
=
IP& + Ie
else if (dv& - dh& > 8)% weak horizontal edge
IP& =
3 IP& + 4
else if - dh& <- 32j%o vertical edge
IP& =
In + In
p& n
else if - dh& <- 8 j% weak vertical edge
}
3!„ +1„
Р& п
The pixels used for prediction the pixel
^ [*' ï]\i=i+s,j=¡+s are: Is = ^ + 1 j]. I = ^ j + 1]. ^ = I[i + 1. j - 1]. Ise = I[i + 1. j + 1]. Iss = I[i + 2. j],
Iee = I[i. j + 2]. IssW = I[i + 2. j - 1].
The gradient of intensity function at the current pixel I is estimated by computing the following quantities:
dfh) I eel ^ ps ^Is
_ ~ Ise\ Is ~ Iss ^\Isw ~ Issw ■
(39)
And the prediction, i. e. concealment, is produced by the following procedure:
if [dVg - dhg > 80 sharp horizontal edge
Ipg le;
(40)
else if (dVg- u,hg
- du <-
80)% sharp vertical edge
else {
I Pg I3 ï
4 '
if {dVg - dhg> 32horizontal edge
In +
r _ Рэ e .
n„ _ .
else if - > 8)% weak horizontal edge 3 • In„ +
I „ =-
Pg e
else if (dy
,g - d^ <- 32)%> vertical edge
In =
+ ^
else if (dVg - dhg <- 8 )% weak vertical edge 3!^ + Ic
Pg s
The pixels used for prediction the pixel
^ ^ ' I=i+3J=y are: Is = I[i + 1. j]. Iw = I[i. j - 1]. Ise = I[i + 1. j + 1]. Isw = I[i + 1. j - 1]. Iss = I[i + 2. j].
Iww = I[i, j - 2]. Isse = I[i + 2. j + 1].
The gradient of intensity function at the current pixel I is estimated by computing the following quantities:
-^шш ^ Is Isw\^\lse Is},
\lw Isw\ ^\IS \lse Isse |* (41)
And the prediction, that is concealment, is produced by the following procedure:
if (dVia - dhiB > 80j°% sharp horizontal edge
I Pis Iw ;
(42)
else if (dVia - dhig < - 80 j°% sharp vertical edge
I Pis I3 ;
else {
t _ 1щ ^ Is 1 Ise Isw . his - 2 + 4 ;
if (dVia - dhiB > 32 j°% horizontal edge
IPis
In + Irn
PlS w
else if ^
- dhiB > 8^% weak horizontal edge
In =-
3Ipis + ^
Pis 4
else if [dVig - dhig <- 32)°% vertical edge
In.„ +1,
I „ =
Pis s
else if (d,, - uh
V~Ui3 nla
Pis 2
- dh._ <- 8^% weak vertical edge
IPis
+1, Pis s
}
After the angular pixels have been predicted (concealed), prediction of the following pixels, P2, P6, P10, P14, is simultaneously performed, based on the known and earlier predicted pixels. All the pixels in the lost block are predicted in the same way.
Results of Computation
In order to demonstrate the advantages of the proposed method of BER calculating, calculations using the new formula of BER were made. These calculations were performed for channel with Ricean fading with different values of K. The results are presented by graphs in Fig. 4 and 5.
In order to demonstrate the advantages of the proposed EC methods, each original video was compared to its resultant (final) videos that were repaired using different EC methods. The results are shown below in three different ways: Histogram, Mean Square Error (MSE) and Peak Signal to Noise Ratio
}
}
2,5 2
Й 1,5 н
<10 3 BER(^), row 10
B
1
0,5 0
2,5 2
0 50 100 150 200 250 300
.ff-factor
10-3 BER(ff), row 15
Й 1,5 E
B
1
0,5 0
2,5 2
Й 1,5 E
0 50 100 150 200 250 300 ^-factor
<10-3 BER(ff), row 68
B
1
0,5 0
0 50 100 150 200 250 300
K-factor
Fig. 4. BER of different channel parameters
(PSNR). Although six different videos were used in the work, this article only presents the graphs of one video and calculation results of all six videos.
The description of the videos that were used for tests of different EC methods is presented below: Video 1
Technical characteristics: Data rate: 631 kbps Total bit rate: 724 kbps Frame rate: 23 frames/s Audio bit rate: 92 kbps Audio channels: 2 (stereo) Audio sample rate: 44 kHz Content & temporary and spatial activity: The fragment presents a shooting of hamburg-
Rician fading channel, <10 3 BER(^) = LCR/R
2,2 1,8
Й 1,4 E B
1
0,6 0,2
0 50 100 150 200 250 300
K-factor
Fig. 5. Average BER of Ricean fading channel
er commercial. A model walking on open catwalk by the sea attracts the attention of tourists. Then she demonstrates a burger and takes a bite of it. The fragment is characterized by medium and slow speed movement, by shifting background, and by the presence of details ranging from large to small.
Video 2
Technical characteristics: Data rate: 371 kbps Total bit rate: 467 kbps Frame rate: 25 frames/s Audio bit rate: 96 kbps Audio channels: 2 (stereo) Audio sample rate: 44 kHz Content & temporary and spatial activity: This fragment from a children's cartoon. In the cartoon there is a constant replacement of the scenery, rain is represented by distinctly significant drops, characters appear and disappear, there is a large change of colors. The fragment is characterized by a lot of fast and slow speed movement, by constantly shifting background, and by the presence of details ranging from large to small. Video 3
Technical characteristics: Data rate: 301 kbps Total bit rate: 397 kbps Frame rate: 29 frames/s Audio bit rate: 96 kbps Audio channels: 2 (stereo) Audio sample rate: 44 kHz Content & temporary and spatial activity: This fragment is a promotional video. There are few characters in the video, changing facial expressions present and the same background is shown from different angles. The fragment is characterized by very little traffic, by mostly unchanging background and by the presence of a range of large and medium-sized details.
Video 4
Technical characteristics: Data rate: 548 kbps Total bit rate: 644 kbps Frame rate: 29 frames/s Audio bit rate: 96 kbps Audio channels: 2 (stereo) Audio sample rate: 44 kHz Content & temporary and spatial activity: The fragment is an artistic and documentary footage of costumed historical scene — an arena surrounded by service buildings. The arena hosts tournaments in real time. The fragment is characterized by constantly shifting background, by a lot of medium speed movement and by the presence of objects and details ranging from large to small. Video 5
Technical characteristics: Data rate: 334 kbps Total bit rate: 382 kbps Frame rate: 18 frames/s Audio bit rate: 47 kbps Audio channels: 2 (stereo) Audio sample rate: 22 kHz Content & temporary and spatial activity: This fragment is from the children's cartoon. The cartoon is characterized by constantly changing scenery and by thrown objects. A child and other characters are in constant moving. The fragment is characterized by a lot of fast and slow speed movement, by constantly shifting background and characters, and by the presence of details ranging from large to small.
Video 6
Technical characteristics: Data rate: 597 kbps Total bit rate: 653 kbps Frame rate: 30 frames/s Audio bit rate: 55 kbps Audio channels: 1 (mono) Audio sample rate: 44 kHz Content & temporary and spatial activity: The fragment is a documentary footage of scenes from a life of little kangaroo with a man. The animal moves all the time, follows the man, tries to repeat all his movements. Man and the little kangaroo play lot. The fragment is characterized by abundance of fast speed movement, by constantly shifting background, by shifting light from bright sun to full shade, and by the presence of details ranging from large to small.
BER comparison. A range of values of K-factor wider than required for the simulation system, K = [0,1, 1, 10, 50, 100, 200, 300], was taken in order to demonstrate the results of BER calculation using the proposed formula. The result of the calculations with all the parameters of the channel is organized in a table consisting of 120 rows and 7 col-
umns. Each column corresponds to one of the values of the K-factor in ascending order. In order to demonstrate the success of the calculations, randomly picked rows were selected from this table and the average values of each column (i. e. the average BER for each value of the K-factor) were counted. The BER values in the selected rows and average BER for different K-factors are presented in Fig. 4 and 5. The graphs have been built on the basis of the rows and the vector of mean values.
As can be seen in the resulting graphs, the lower K-factor leads to the higher BER, and the higher K-factor leads to the lower BER. Thus, for smaller values of K the BER closes to Rayleigh fading, and for larger values of K the BER tends to zero, i. e., closes to Gauss fading.
Error Concealment comparison. The results of the comparison between suggested EC models and existing models, performed by Mean Square Error (MSE) and Peak Signal to Noise Ratio (PSNR), are represented by:
1) the table of average MSE relations (ratios) that were calculated by following formula (Tabl. 1 and 2):
average MSE relation = average MSE of tested {existed } EC model
average MSE of my Symmetric CALIC model
2) histogram based on MSE relations (ratios) that were calculated by following formula:
MSE relation array = MSE array of tested (existed} EC model
MSE array of my Symmetric CALIC model' ( )
3) graph of MSE of all methods;
4) graph of PSNR of all methods.
Existing models: Symmetric Intra Prediction using System of Linear Equations (SISLE), Weighted
■ Table 1. MSE average Ratios MSEaveRelations
1x5 double
1 г 3 4 5
2.45Î5 г. ,1365 5.0872 12627 22933
■ Table 2. MSE average Ratio: Balanced Calculation 50-50 % method compared to SCALIC method
M SEaveRea Iti on sB a I a n c ed vsSym C ALJC
lid double
1 2 3
0.4073
50 0
50 0
Symmetric CALIC vs Symmetric ISLE
1
0 2 4 6 8 10 12 Symmetric CALIC vs Weighted Averaging
100 50 0
0123456789 10 Symmetric CALIC vs ISLE
100 50 0
0 5 10 15 20 25 30 35 40 45 50 Symmetric CALIC vs Standard CALIC
-нШ Шгг
0
100 50 0
1 2 3 4 5
Symmetric CALIC vs Partial WA
0
10 12 14
Fig. 6. Histogram of MSE ratios: SCALIC method compared to existing methods
a) 0,45
0,4
0,35
0,3
H 0,25 S
^ 0,2 0,15 0,1 0,05 0
0 20 40 60 80
Frames
100 120
b) 200 190 180 170
Ц 160
PS150
140 130 120 110
ШшйЫВДи
MIfP
20 40 60 80
Frames
-*- Sym. CALIC SISLE
100 120
Std. CALIC Part. WA
WA ISLE
Fig. 7. SCALIC method compared to existing methods: a — MSE comparison; b — PSNR comparison
Symmetric CALIC vs Balanced 50 % - 50 %
R
70 60 50 40 30 20 10 0
Fig. 8. Histogram of MSE ratios: Balanced Calculation method compared to SCALIC method
Averaging (WA), Intra Prediction using System of Linear Equations (ISLE), Standard CALIC (Std. CALIC), Partial Weighted Averaging (Part. WA).
It should be noted that the work was carried out with the I-frames. The impact of the proposed method on other types of frames is a topic for further study and work.
In the beginning, the results of comparison between SCALIC and other existed methods are presented. Then the results of comparison between other proposed method (Balanced Percentage Calculation) and SCALIC are presented.
Results of Comparison (for video fragment 5).
The SCALIC method compared to different existing methods is shown in Fig. 6 and 7 in the form of histograms and in graphical form, respectively.
The Balanced Calculation 50-50 % method compared to SCALIC method is presented in Fig. 8 and 9
Table 3. SCALIC compared to different existing models
Average MSE per model per video SCALIC vs
SISLE WA ISLE Std. CALIC Part. WA
Video fragment 1 4,1567 3,2985 12,2934 1,3119 4,2980
Video fragment 2 3,0673 2,6818 6,9647 1,1516 3,0214
Video fragment 3 1,5805 1,5285 2,2497 1,0271 1,6583
Video fragment 4 2,3352 2,1787 3,6461 1,0390 2,2704
Video fragment 5 2,4595 2,1365 5,0872 1,2627 2,2938
Video fragment 6 4,5981 3,5052 10,0879 1,0622 4,2013
Average MSE per model 3,0329 2,5549 6,7215 1,1424 2,9572
7
0
1
2
3
4
5
6
6
0
a)
20 40 60 80
Frames
100 120
20
40 60 80
Frames
100 120
-*- Sym. CALIC - Balanced 50 - 50 %
Fig. 9. Balanced Calculation 50-50 % method compared to SCALIC method: a — MSE comparison; b — PSNR comparison
Table 4. Balanced Calculation 50-50 % compared to SCALIC
Average MSE per model per video Balanced 50-50 % vs SCALIC
Video fragment 1 0,4221
Video fragment 2 0,3854
Video fragment 3 0,5720
Video fragment 4 0,3953
Video fragment 5 0,4073
Video fragment 6 0,2621
Average MSE per model 0,4888
in the form of histogram and graphically, respectively.
In order to show the general results, the average values of MSE of all used videos were calculated. The corresponding comparison between all the proposed methods is summarized in Tabl. 3 and 4.
For a better view of the effectiveness of EC methods we will present their ranking in decreasing order from the most effective method to the least effective method (it should be noted that all these methods, including the least effective, are working):
1) Balanced Percentage Calculation 50-50 %;
2) Symmetrical CALIC;
3) Standard CALIC;
4) Weighted Averaging;
5) Partial Weighted Averaging;
6) Symmetric ISLE;
7) ISLE.
Summary
BER. The proposed method of BER calculating is versatile and provides the receiving of BER for a channel with any of the three common types of fading. This makes it possible to use all the channel's available resources for one species fading (Ricean), thus increasing the accuracy of the necessary calculations and predictions. That, in turn, improves the accuracy of filling errors.
Based on the results shown in the previous Section, we can see evidence of the validity of the suggested method. Thus, for different values of K-factor, a BER for the channel with different types of fading is achieved.
Error Concealment. Based on the results of the simulation in the previous Section, we can conclude that the symmetrical EC, performed from the four corners (four sides and each side is used in two of calculations) of the lost block, is very perspective. This concealment uses twice as much or more of the existing information than asymmetrical (with one corner and two sides and other) processes of EC. The suggested method (SCALIC) achieves on average an improvement in EC ranging from 14 to 670 % compared to various asymmetric methods. This improvement is very significant.
These results are logical and consistent due to the fact that using of all the information surrounding the lost block (two rows and two columns) provides better results than use of only half of the information (in the asymmetrical methods that use is only one row and one column).
However, additional method of EC that is proposed in this work, called Balanced Percentage Calculation, provides results two times better on average than SCALIC. This indicates that this method is more efficient and attention should be paid to the development of algorithm for it.
It should be noted that the proposed methods were compared to asymmetric methods (that use the top row and the left column — CALIC and ISLE) in order to reference this study with previous studies and in order to show that there are undoubted benefits in using symmetric techniques.
0
Practice Recommendations
BER. In future research it is necessary to develop
E
an algorithm for the transition from —— in the clas-
N0
sical formulas for BER calculating to the parameters (arguments) of the proposed formula to enable rapid performance of necessary calculations and comparisons.
In addition, it makes sense to study the possibility of constructing an algorithm to determine the optimal value of the K-factor for each of the proposed (existing) communication channel.
Error Concealment. In future research, more attention should be paid to different symmetrical EC methods that might be very successful and show much better results than the existing asymmetrical ways. This confirms not only the SCALIC, proposed in this work, but also ISLE and Symmetrical ISLE methods that were used for comparison: symmetrical prediction gives on average results that are twice or more better than unsymmetrical. The SCALIC itself compared to the Std. CALIC gives a 14 % aver-
References
1. Wu X.. and Memon N. CALIC — a Context Based Adaptive Lossless Image Codec. Proc. Int. Conf. on Acoust, Speech, Signal Processing. Atlanta. GA. May 1996. pp. 1891-1894.
2. Zhao X.. Lu H.. Chen C. W.. and Wu J. Adaptive Hybrid Digital-analog Video Transmission in Wireless Fading Channel. IEEE Trans. on Circuits and Systems for Video Technology, 2016. no. 6. pp. 1215-1223.
3. Shen X.. Tang L.. and Zhang X. A Video Transmission Scheme Based on Scalable Video Coding over MIMO Wireless Systems. Intern. Conf. on Wireless Communication and Sensor Network. 2014. 6 p.
4. Purandare R. G.. Kshirsagar S. P.. Koli S. M.. and Go-hokar V. V. Impact of Bit Error on Video Transmission over Wireless Networks and Error Resiliency. Intern. Conf. on Image Information Processing (ICIIP). 2011. 6 p.
5. Bocus M. Z.. and Coon J. P. Rate-Optimization for Scalable Video Transmission over Wireless Networks. IEEE Press. 2014. 287 p.
6. Hadar O.. Huber M.. and Huber R. Hybrid Error Concealment with Automatic Error Detection for Trans-
age improvement in the results of EC, which is undoubtedly a good result, although more research is needed to improve it.
However, it is also worth paying attention to the fact that the full information, surrounding the lost block, is not always available. In some cases only half of the data (two sides) is available that leads to slightly worse results. This can be seen by comparison between WA and Part. WA, wherein the information from two sides is used: Part. WA shows results that on average are worse only 1,157 times than WA. This was revealed in the process of this research and it is important topic for future research.
Another topic for future research is Balanced Percentage Calculation for replacement of pixels of lost block. As can be seen from the results of the previous Section, this method produces results twice better on average than the SCALIC, thereby demonstrating its high efficiency. Therefore, future research is needed to develop an algorithm to determine the optimal percentage of this calculation.
mitted MPEG-2 Video Streams over Wireless Communication Network. Intern. Conf. on Information Technology: Research and Education, 2006, 6 p.
7. Hadar O., Huber M., Huber R., and Greenberg S. New Hybrid Error Concealment for Digital Compressed Video. EURASIP Journal on Applied Signal Processing, 2005, no. 6, pp. 1821-1833.
8. Mazataud C. Error Concealment for H.264 Video Transmission. A Thesis Presented to the Academic Faculty, In Partial Fulfillment of the Requirements for the Degree of Master of Science in the School of Electrical and Computer Engineering, Georgia Institute of Technology, 2009. 94 p.
9. Patel D., and Patel J. Hybrid Spatio-Temporal Error Concealment Technique for Image/Video Transmission. IEEE Proc., 2014, no. 5, pp. 814-823.
10. Lee Y. H., and Wu T. H. Video Error Concealment Using Progressive Block Matching Algorithm with Euclidean Distance. Eighth Intern. Conf. on Intelligent Information Hiding and Multimedia Signal Processing, 2012, 6 p.
УДК 621.39
doi:10.15217/issn1684-8853.2017.2.67
Оптимизация маскирования ошибок на основе анализа типов затухания
Часть 2: Модифицированные и новые модели маскирования ошибок видеосигналов.
Практическое моделирование и его результаты
Офер Хадара, Phd, профессор Ирина Бронфман6, MSc студент
Натан Блаунштейна>б, доктор физ.-мат. наук, профессор аНегевский университет им. Бен-Гуриона, г. Беэр-Шева, Израиль бИерусалимский технологический институт, Иерусалим, Израиль
Цель исследования: представить наиболее эффективный метод расчетов параметров канала с затуханием и разработать методы достижения лучшего и более эффективного выполнения замещения ошибок, что повысит качество видео после прохождения через канал с затуханием. Методы: исследованы влияние затухания на канал связи при помощи изучения распределений Гаусса, Рейли и Райса, а также существующие методы прогнозирования и замещения ошибок и их влияние на качество видео после выхода из канала с затуханием. Результаты: показано, что распределение Райса более широкое и охватывает другие виды распределения — Гаусса (идеальный канал) и Рейли (канал с затуханием), поэтому, именно это распределение было использовано для тестирования практических случаев, возникающих в видеоканале. По теме замещения ошибок проведена оптимизация метода CALIC, названная Симметричный CALIC (Symmetric CALIC). Реализовано сравнение данной оптимизации с оригинальным методом CALIC и с другими методами и определено, что предлагаемая оптимизация показывает лучшие результаты, чем все использованные для сравнения методы. Предложен новый метод замещения ошибок, названный Сбалансированным Процентарным Расчетом (Balanced Percentage Calculation), в сравнении показавший в среднем в два раза лучшие результаты, чем Симметричный CALIC, и намного лучшие результаты, чем остальные использованные для сравнения методы. Две темы объединены таким образом, что затухание повлияло на появление ошибок в видеофайле, которые были исправлены при помощи предложенных методов замещения ошибок. Все практические тесты и сравнения проведены в MatLab. Практическая значимость: предложенный способ расчета параметров канала с затуханием позволяет выполнять расчеты для любых видов каналов, что значительно облегчает работу с каналами в общем и с необходимыми для них расчетами в частности. Предложенные оптимизация существующего метода замещения ошибок и новый метод замещения ошибок позволяют получать видео более высокого качества после прохождения через канал с затуханием.
Ключевые слова — распределение Райса, затухание Райса, скорость ошибочных битов, BER, замещение ошибок, коэффициент уровня пересечения, средняя продолжительность затухания, CALIC, симметричный CALIC, сбалансированный процентар-ный расчет, средняя взвешенность, внутрикадровое прогнозирование с помощью системы линейных уравнений.
