![](bg9.png)
Table 2: Quality assessement, using PSNR.
Video name O vs. TC O vs. TM TC vs. TM
Documentary 33.3664 33.7969 32.5094
Music clip 40.0416 37.1115 36.2243
Basketball 34.0789 35.9083 32.5832
Game screen rec. 42.0799 43.0613 43.4582
four, our transmoder gives better result compared to
the transcoded version of the original video (Docu-
mentary, Basketball and Game screen record). For
the documentary video, the PSNR increases, which
means that the video quality is better, while necessary
bitrate was reduced of about 6.04%. For the game
screen record: our transmoder saves 8.6% in bitrate
while increasing the PSNR from 42.0799 to 43.0613.
As a short conclusion, we can say that our trans-
moder can provide better results both in bitrate and
PSNR terms for most of the test videos.
5.4 Real Time Processing
Our first tests, presented in Table 3 reveal that real
time processing is possible with videos assuming
quite little spatial resolutions (240p). Using this spa-
tial resolution, a 29 framerate is possible for the mu-
sic clip. This framerate isn’t reachable for all tested
videos, as an example the documentary framerate is
only about 7 pictures per second. In order to over-
come this limitation, we first optimize the whole
architecture of the software, especially by adding
threading support in appropriate source code parts.
Another way of increasing performance, especially
in the image processing world is to use the GPU.
This aspect has been brought to our program using
OpenCV GPU related functions.
Table 3: Transmoding time (second) vs. spatial resolution.
Video name 240p 480p 720p
Documentary 65.52 142.28 295.73
Music clip 14.50 54.27 111.84
Basketball 56.11 128.45 281.19
Game screen record 18.03 113.08 261.17
6 CONCLUSION AND FUTURE
WORKS
In this paper, we presented a new kind of video en-
coding system called transmoder. The video stream
is splitted into regions that are encoded using sev-
eral modalities depending on the regions character-
istics. We propose an overall system architecture for
transmoding by employing two modalities, pixel and
vector encoding. We first split each frame into re-
gion using an edge detector and then determine the
more relevant encoder for each region. Where the
resulted output of biomodal video stream is a com-
bination of vector and pixel frames. We tested our
approach against several real life videos and perfor-
mance analysis shows that our approach significantly
outperforms state-of-the-art encoder for a large ma-
jority of testbed videos. These tests present a reduc-
tion of the necessary bitrate up to 8%. Future research
will allow us to optimize even more our system, us-
ing for example GPU optimized algorithms. In the
frame of our research on cloud gaming systems, we
aim to integrate this multimodal coding scheme in the
realtime rendering chain. We therefore adapt our ap-
proach for distributed processing in a cloud architec-
ture.
REFERENCES
Chaudhury, S., Mathur, M., Khandelia, A., Tripathi, S.,
Lall, B., Roy, S. D., and Gorecha, S. (2011). Sys-
tem and method for object based parametric video
coding. U.S. Classification: 375/240.16; 375/240.25;
375/E07.027; 375/E07.076.
Felzenszwalb, P. F. and Huttenlocher, D. P. (2004). Effi-
cient graph-based image segmentation. International
Journal of Computer Vision, 59(2):167–181.
Ferrari, L., Sankar, P., and Sklansky, J. (1984). Minimal
rectangular partitions of digitized blobs. Computer Vi-
sion, Graphics, and Image Processing, 28(1):58–71.
Heath, M., Sarkar, S., Sanocki, T., and Bowyer, K. (1996).
Comparison of edge detectors: a methodology and ini-
tial study. In Proceedings CVPR ’96, 1996 IEEE Com-
puter Society Conference on Computer Vision and
Pattern Recognition, 1996, pages 143–148.
Khandelia, A., Gorecha, S., Lall, B., Chaudhury, S., and
Mathur, M. (2008). Parametric video compression
scheme using AR based texture synthesis. In Sixth
Indian Conference on Computer Vision, Graphics Im-
age Processing, 2008. ICVGIP ’08, pages 219 –225.
Lai, Y.-K., Hu, S.-M., and Martin, R. R. (2009). Auto-
matic and topology-preserving gradient mesh gener-
ation for image vectorization. ACM Trans. Graph.,
28(3):85:185:8.
Maini Raman, D. H. A. (2009). Study and comparison
of various image edge detection techniques. Interna-
tional Journal of Image Processing.
Orzan, A., Bousseau, A., Winnemller, H., Barla, P., Thollot,
J., and Salesin, D. (2008). Diffusion curves: a vec-
tor representation for smooth-shaded images. ACM
Trans. Graph., 27(3):92:192:8.
Sun, J., Liang, L., Wen, F., and Shum, H.-Y. (2007). Im-
age vectorization using optimized gradient meshes. In
ACM SIGGRAPH 2007 papers, SIGGRAPH ’07, New
York, NY, USA. ACM.
ICEIS2014-16thInternationalConferenceonEnterpriseInformationSystems
36