Post on 12-Feb-2016
description
Abhijith JagannathEE5359 – Project Proposal
*Analysis of Adaptive GOP Algorithms forEfficient HEVC Compression
*Topics
Basics of video codingGOP and GOP structuresWhy do we need adaptive GOP algorithm?Already proposed methodsImplementation Schematic on HEVC.Scope of the projectReferences
*Basics of Video Coding
*Coding Continued..
* GOP and GOP Structures
GOP - Group Of Pictures ( Video Frames ). I-Intra Frame : P – Predicted frame : B –
Bidirectional predicted frame Order in which different types of frames are
placed makes a GOP structure A classic GOP structure is shown below
*Other GOP Structures
* Why Adaptive GOP?
Fixed GOP structures prevent encoders from adapting to temporal variations in video sequences [6].
The transitions between scenes are the regions, where static GOP structures achieved poor performance[4].
Adaptively changing the sizes of GOP structure according to the temporal characteristics of a video sequences improves the coding efficiency [6] [5].
*Existing Adaptive Algorithms
Adaptive GOP structure based on the positions of video cuts [4].
Adaptive GOP based on Sum of Absolute Motion Vectors (SAMV) [6].
Scene Change Detection (SCD) based on motion and residual information [5].
*Implementing Adaptive GOP Structure in HEVC
Modified from HEVC block diagram [1]
*Static and Dynamic GOP
To make the GOP structure dynamically adaptive for any video sequence, there is a big challenge of changing the code in more than 100 places in the HEVC video codec.
When a known video is encoded, GOP structure can be easily determined and hard coded.
Hard coding of GOP for a given particular video can be called as static adapted GOP structure
As a proof of concept and to analyze adaptive GOP structure, static method is used.
*Creating test sequences
Custom test sequences are created by extracting frames from the standard sequences.
The tools used are ffmpeg [8], YUV Player [10] and YUV Tools [9]
Frames taken from standard video sequence
* Test Sequence example Custom test sequences is created with known
pictures. For the example, 4 pictures are selected and
repeated for 4 frames t0 make a HD video of 12 frames.
Hence we can change the GOP structure after 4 frames accordingly.
Frames of custom video sequence
* Test Analysis Details
All the test sequences are made high definition videos.
16 frames are coded for custom made sequence.
12 frames are codes for standard sequence. Each sequence is codded as single frame per
second. HM 10.0 [7] is used to encode the test
sequence.
* Results
encode
r_ran
domacc
ess_m
ain.cf
g
encode
r_lowdel
ay_m
ain.cf
g
encod
er_low
delay
_P_main
.cfg
Adap
tive_B
Adap
tive_P
35
36
37
38Bitrate of custom sequence in kbps
encod
er_ran
domacc
ess_m
ain.cf
g
encode
r_lowdel
ay_m
ain.cf
g
encod
er_low
delay
_P_main
.cfg
Adap
tive_B
Adapt
ive_P
4444.144.244.344.444.544.644.744.8
Avg PSNR of custom sequence in dB
* Results
encod
er_ran
domacc
ess_m
ain.cf
g
encode
r_low
delay
_main
.cfg
encod
er_low
delay
_P_main
.cfg
Adapt
ive_B
Adap
tive_P
6264666870
Bitrate of standard sequence in kbps
encoder_randomaccess_main.cfg
encoder_lowdelay_main.cfg
encoder_lowdelay_P_main.cfg
Adaptive_B Adaptive_P40.8
40.9
41
41.1
41.2
41.3
41.4
41.5
41.6
41.7
Avg PSNR of standard sequence in dB
* Results
0400800
1200
Encoding time in seconds - Custom Sequence
encode
r_ran
domacc
ess_m
ain.cf
g
encod
er_low
delay
_main
.cfg
encode
r_lowdel
ay_P_
main.cf
g
Adap
tive_B
Adap
tive_P
0100200300400500600700800
Encoding Time in seconds - Standard Sequence
*Moving Forward Analyzing and recognition of a unique GOP
pattern that can give best results.
When the standard pattern is recognized, the project can be taken forward to implement the dynamic GOP structure.
*References1. G.J. Sullivan et al, “Overview of high efficiency video coding (HEVC)
standard”, IEEE Trans. CSVT, vol. 22, pp, 1649 - 1668, Dec. 20122. I.E.Richardson, “The H.264 advanced video compression standard”,
Wiley, Second edition, 2010.3. P. Anjanappa, “Performance analysis and implementation of mode
dependent DCT/DST in H.264/AVC”, M.S. Thesis, EE Dept., UTA, Arlington, Texas, 2012. http://www-ee.uta.edu/dip/Courses/EE5359/Priya_Thesis_Final.pdf
4. Y. Yokoyama, “Adaptive GOP structure selection for real-time mpeg-2 video encoding,” IEEE international conference on image processing, vol.2, pp. 832-835, Sep. 2000.
5. L. Krulikovska. “A novel method of adaptive GOP structure based on the positions of video cuts ” ELMAR, 2011 Proceedings.
6. Jun-Ren Ding, Ji-Kun Lin and Jar-Ferr Yang, “Motion-based adaptive GOP algorithms for efficient H.264/AVC compression,” IEEE Transactions on CSVT, vol 22 , Issue: 1, pp.28-42, Jan 2012.
7. HM Software: https://hevc.hhi.fraunhofer.de/svn/svn_HEVCSoftware/ 8. FFMPEG website : http://ffmpeg.org/ 9. YUV Tools: http://www.sunrayimage.com/yuvtools.html 10.YUV Player: http://www.yuvplayer.com/