Page 216 - ITU Kaleidoscope 2016

2016 ITU Kaleidoscope Academic Conference

Table 2. List of parameters in HDR [7] compute it exhibits imprecise measurement when used to
measure the quality of streamed video over wireless network.
Parameters Value Definition Since the packet loss in the wireless mobile network cannot
be neglected, more complex metrics like A-PSNR, VQM,
T 35 sec Target Buffering and different types of SSIM are employed for the evaluation
Time of system.
d 60 sec Time Period
estimating 6. RESULTS AND DISCUSSIONS
connection
throughput
N 2, N 1, Z, P 1, 0.25, 0.5, 1, Factors of 6.1 Peak Signal to Noise Ratio (PSNR) Measurement

P 2 1.5, 2 membership The PSNR metric was evaluated offline on the data
functions generated by a live stream during the experimentation for the
three implemented algorithms. The proposed ABBA
5. IMPLEMENTATION ENVIRONMENT performs better as it well predicts the future trend and allows

Different video resolutions namely SQCIF, QCIF, CIF, the video content as per the network conditions thereby
minimizing the mean square error in the decoded frame at
QVGA, and VGA were used for the transcoding of the input the receiver. Table 3 lists the PSNR values corresponding to
video at the server side for every streaming instance. The the ABBA (proposed), BSR, and HDR algorithms. The
frames per second (fps) designated for the streaming are 10, ABBA algorithm exhibits a higher average PSNR, which is
15, 25, 30 and 35 respectively with the default value set to 21% and 12% higher than the buffer based (BSR) algorithm
25 fps. Java programming environment was used as it is and Heuristic based (HDR) algorithm respectively.
platform independent and supports VLCJ (VLC for Java).
The VLCJ is an open source framework that is used for video Table 3. Comparison of PSNR values
streaming that enables the media content to get embedded in
a Java Swing. Since this platform is completely open source # ABBA HDR BSR # ABBA HDR BSR
there are many customizable options available that are (dB) (dB) (dB) (dB) (dB) (dB)
deployed for obtaining the media statistics. Synchronization 1 36.01 24.26 24.50 11 35.03 27.86 31.54
of multithreads is carried out for communication between the 2 36.35 32.26 24.25 12 35.03 27.69 31.07
client and server for transfer of data. Transcoding of input 3 35.11 32.25 36.10 13 32.04 35.69 31.21
stream i.e., the process of converting media object from one 4 34.11 35.10 29.28 14 36.35 35.97 31.92
configuration to another, allows to switch between various 5 35.11 29.28 28.32 15 36.35 25.21 25.73
resolutions at the server side. Dshow [17] is the API that is 6 33.11 28.32 31.22 16 37.35 35.54 27.69
used to capture the video and process it for streaming in the 7 35.125 24.50 31.38 17 34.37 27.69 27.75
appropriate format. 8 35.116 34.38 27.54 18 36.53 35.02 29.85
The wireless network for experimental set up was 9 35.32 34.22 27.46 19 37.37 35.62 31.98
established through Airtel 4G LTE-TD Hotspot [6] that 10 35.066 27.29 31.97 20 31.24 36.19 32.04
demonstrated an average of 3.7 Mbps in downlink during Average(dB) 35.10 31.37 29.13
real time testing although it is intended to support more than
8 Mbps as specified by the service provider. For case study 6.2 Structural Similarity (SSIM) Measurement
the end-to-end link bandwidth was estimated with the help
of an online tool Speedof.Me [18]. The client and server were The SSIM index was computed for the three algorithms
implemented in Lenovo idea pad laptop which has Intel Core (ABBA, BSR, and HDR), and the proposed ABBA
i5 processor having 4 GB RAM and Windows 7 Professional algorithm exhibited a higher value with an average of 0.9541
64 bit operating system. The streaming of video was which is 9% higher than the buffer based algorithm 7% better
implemented on top of the HTTP with UDP as its underlying than heuristic based algorithm.
transport protocol. The SSIM index on few decoded consecutive frames at
receiver corresponding to ABBA, BSR, and HDR algorithms
5.1 Performance Evaluation Parameters is plotted in Figure 5. The higher SSIM index for ABBA
algorithm is a reward for the perceived video quality as there
is need for the perseverance of luminance and contrast
Since in our experimental set up original video sequence was factors that are influenced by the distortions.
readily available, the Full Reference (FR) metrics were
employed to evaluate the system performance. Further, FR
metrics provide the most accurate result as it is computed
with direct reference to the original sequence. The
commonly used FR parameters are: Peak Signal to Noise
Ratio (PSNR), Structural Similarity (SSIM) index, Video
Quality metric (VQM), Aligned PSNR (A-PSNR) etc.
Although the conventional PSNR is relatively simple to

– 198 –