Selection of intra prediction modes for intra frame

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
__________________________________________________________________________________________
Volume: 02 Issue: 12 | Dec-2013, Available @ http://www.ijret.org 67
SELECTION OF INTRA PREDICTION MODES FOR INTRA FRAME
CODING IN ADVANCED VIDEO CODING STANDARD
Manjanaik.N1
, Manjunath.R2
1
Electronics and Communication Engineering, Jain University, Bangalore, India.
2
Senior Domain Specialist, Philips Company, Bangalore, India.
manjubdt2009@gmail.com, manju_r_99@yahoo.com
Abstract
This paper proposes selection of Intra prediction modes for Intra frame coding in Advanced Video Coding Standard using Matlab.
The proposed algorithm selects prediction modes for intra frame coding. There are nine prediction modes are there to predict the
intra frame in AVC using Intra prediction,but all the prediction modes are not required for all the applications. Intra prediction is the
first process of advanced video coding standard. It predicts a macro block by referring to its previous macro blocks to reduce spatial
redundancy,appling all the prediction modes to predict intra frame it leads to more computational complexity is increased at the
encoder of AVC. In the proposed algoriyhm, applied all the prediction modes(0-8) for prediction of intra frame but only few modes
such as mode0, mode1, mode2,mode4,mode6 gives good PSNR, high comprssion ratio and low bit rate. Out of these modes mode2
gives good PSNR, compression ratio and redced bit rate, mode5, mode7 and mode8 gives lower PSNR, low compression ratio and
increased bitrate compared to mode0,mode1, mode2, mode4 and mode6. The simulation results are presented using Matlab. The
PSNR , compressed ratio and bit rate achived for different quantization parameters of mother daughter frames , foreman frames was
presented.
Keywords: AVC, PSNR, CAVLC, Macroblock, Prediction modes.
------------------------------------------------------------------------***----------------------------------------------------------------------
1. INTRODUCTION
Compression is the basic process of reduceing the size of data
in order to save storge space and transmission band
width.compression consist of removing redundancies
(spatial,spectra and temporal) and encoding the true
information in the form of appropriate to suite for applications.
There are two compression techniques ie. Lossless and lossy
techniques. In lossless the reconstured image after comprssion
is identical to original image, this method achieve maximum
compression ratio.To compress data ,it is important to
recognize redundancies in data in the form of coding
redundancy, inter- pixel redundancy, and psycho-visual
redundancy. Data redundancies occur when unnecessary data is
used to represent source information. compression is achived
when one or more of these types of redundancies are
reduced.In lossy the reconstructed image is not identical to the
original image ie there is loss in information.lossy method is
capble of achiving a high comprssion compare to lossless
method. Image compression/video coding is impotant in
indusrial imaging, commercial and academic applications.
Image /video coding plays a important role in multimedia.
H.264/AVC is the latest video coding standard jointly
developed by Joint video team which is organized by two
international standards bodies ie the Internatonal
Telecommunication Union-Telecommunications sector (ITU-
T) and International Organization for Standardization/
International Electro-technical Comission (ISO/IEC). This
standard consists of various adavanced features(Intra prediction
unit, integer transform, variable block mation estimation,
entropy encoding ,deblocking filter and coding tools.due to
these features this standard achieves greater compression
without sacrificing on video quality. Intra prediction unit is the
first unit of H.264 standard. H.264 standard support intra
prediction for diffent size ie 16x16 macro block as whole or
4x4 sub blocks.there are seventeen prediction modes for a
macroblock, nine prediction modes for 16 4x4 luma sub
macroblocks, four modes for a 16x16 luma macro block and
four modes for two each two chorma 8x8 blocks. All possible
combinations of intra prediction modes are used it lead to
increase computational complexity of H.264 encoder and it is
difficult to use for real time applications.All the modes are not
relavant to all the applications, so that proposed algorithm is
implemented for nine prediction mode.out of these mode few
modes gives good PSNR, high compression ratio and low bit
rate. These modes can use for suitable multi media
applications.
2. METHODOLOGY
A raw recorded video in yuv format is used as an input file.
The video is then decomposed into frames(I-frame). Each Intra
frame is processed in terms of 4x4 sub macro block. For each
Intra frame processing is done in order to get good video

__________________________________________________________________________________________
quality, high compression ratio and low bitrate as compared to
previous video coding compression standards. Each Intra
frame processing includes frame conversion i.e. RGB to
YCBCR format down sampling i.e. (4:4:4 to 4:2:0), 4x4 sub
macro block segmentation, all prediction modes (vertical,
horizontal,DC, diagonal down left, horizontal down, diagonal
down right, vertical left, horizontal up, vertical right etc)
integer transformation, quantization, scaling and entropy
encoding(cavlc). The video is reconstructed by inverse
quantization, inverse transformation and deblocking filter.
3. BLOCK DIAGRAM
The block diagram of H.264 encoder for selection of Intra
prediction modes is shown in fig.1. The h.264 encoder block
consists of integer transform, quantization context adaptive
variable length coding, inverse quantization, invese
transformation, intra prediction unit and deblocking filter.
Fig.1 Block diagram of H.264 encoder
The block diagram of H.264/AVC encoder includes two
dataflow paths, a forward path and a reconstruction path. An
input frame is given for encoding. Every frame is processed in
terms of a Macroblock (MB) of size 16x16 pixels. Each
macroblock is further sub divided into 4x4 sub macroblock.
Each 4x4 sub macroblock is encoded in intra prediction modes.
A prediction macroblock P is formed based on a reconstructed
block. In intra mode, P is formed from samples in the current
block is based on previously reconstructed block. The
prediction P is subtracted from the current macroblock to
produce a residual or difference macroblock. This is
transformed using integer transform and quantized using
quantization block to give quantized transform coefficients.
These coefficients are reordered and entropy encoded using
context adaptive variable length coding (CAVLC) and the
compressed bit stream is transmitted over a band-limited serial
transmission channel . In the reconstruction path the quantized
macroblock coefficients are decoded to reconstruct a frame for
encoding of other macroblocks. The quantized coefficients are
inverse quantized and inverse transformed to produce a
difference macroblock. The prediction macroblock P is added
to difference maroblock to create a reconstructed macroblock
after a de-blocking filter, which improves the quality of the
reconstructed frame[1-2].
4. INTRA PREDICTION
The H.264/AVC intra prediction unit achieves higher
compression ratio and image quality compared with preivious
standard(JPEG2000).The H.264 support different block sizes, it
supports 4x4 and 16x16 block sizes for base line, main and
extended profiles and 8x8 block size for high profile. There are
nine prediction modes for 4x4 blocks, four for 16x16 blocks
and and two for 8x8 blocks. All the prediction pixels are
calculated based on the the reconstructed pixels of previously
encoded neighbouring blocks. The prediction of 4x4 blocks is
predicted based on the previously reconstructed pixels labelled
(A-M) shown in Fig.2 the pixels (A-M) are reconstructed
previously and consider as reference pixels for current block.
The pixels labeled (a-m) are prediction pixels.
Fig. 2. labeling of 4x4 prediction samples
Each 4x4 sub macroblock is predicted using eight directional
prediction modes and one DC mode.The directional prediction
modes are vertical, horizontal, diagonal down left, horizontal
down, diagonal down right, vertical left, horizontal up, vertical
right.For directional modes the predicted samples are formed
from a weighted average of the perdiction samples A-M.For
DC mode the predicted samples are formed by mean of
samples A-D and I-L.The encoder select prediction mode for
each 4x4 sub macroblock.The selection of best prediction mode
is obtained by minimizing the residual encoded block and its
prediction[3-5]. The fig.3 shows the intra prediction modes.
Fig.3. 4x4 intra prediction modes
M A B C D E F G H
I a b c d
J e f g h
K i j k l
L m n o p

__________________________________________________________________________________________
The equations of 4x4 sub macroblock for few prediction modes
are:
Vertical
a=e=i=m=A
b=f=j=n=B
c=g=k=o=C
d=h=l=p=D
Horizontal
a=b=c=d=I
e=f=g=h=J
i=j=k=l=K,
m=n=o=p=L
DC
Mean of A-D and I-L,
for horizontal prediction sum of (A-D) and vertical prediction
sum of (I-L).
Diagonal Down Left
a=(A+2B+C+2)>>2
b=e=(B+2C+D+2)>>2
c=f=i=(C+2D+E+2)>>2
d=g=j=m=(D+2E+F+2)>>2
h=k=n=(E+2F+G+2)>>2
i=o=(F+2G+H+2)>>2
p=(G+3H+2)>>2.
5. STEPS OF IMPLEMENTATION
The following steps are to required to selection of intra
prediction modes for intra frame coding in AVC.
• A recorded video in YUV CIF format as input file
• This file is decomposed into Intra frames (1-30)
• Each I-frame is divided into 16x16 macroblocks
• Eac h 16x16 macroblock further sub divided into 4x4 sub
macroblocks
• A first 4x4 sub block is processed directly without using
previously reconstructed block followed by integer
transform, quantization, entropy encoding (cavlc) at
encoder and reverse process at reconstruction path.
• Reconstruct a 4x4 subblock using inverse process (at
reconstruction path)
• Obtain residual block by subtracting next 4x4 sub block
with previously reconstructed sub block
• Residual of 4x4 sub block is integer transformed,
quantized and entropy encoded at encoder and reverse
process at reconstuction path.
• Finally measure PSNR, Compression ratio and Bit rate of
intra frames with different QP for all prediction modes
6. IMPLEMENTAION
The proposed work is carried out using Matlab. The input is
yuv sequences CIF format video file. For test purpose mother-
daughter frame of resolution 352x288 and also other frames
(foreman, news, etc) also taken, A Matlab program is written,
which reads the yuv video file, extracts Intra frames. The next
process involves reading a true RGB colour frame and convert
into ycbcr format down sampling (4:4:4 to 4:2:0) to reduce bits
of intra frame.ycbcr intra frame is divided into 4x4sub block is
processed directly by following usual procedure of forward
path of H.264 encoder and reconstruct the processed block
which serves as reference to the next sub block using basic
reverse process in reconstruction path of h.264 encoder. Apply
all the prediction modes to each sub block followed by,
quantization, context adaptive variable length coder (CAVLC)
to get compressed bit. At H.264 encoder in the reconstruction
path, perfom reverse process to get reconstruct image and
finally measure quality picture (PSNR).
7. RESULTS AND DISCUSSION
The proposed method, selection of intra prediction modes and
best prediction mode for intra frame is done using Matlab. The
test sequences are coded with only intra frames with frame rate
25 frames per second.The results obtained for the test yuv
sequences in CIF format, five forman cif frames for
quantization parameter 30 and ten mother-daughter cif frames
for quantization parameter 35. Other quantization
parametes[10, 25, 40, 45,51] also chosen (not shown in table).
Table-1
PSNR, Compression ratio, Bit-rate, of foreman cif frames (1-5) with QP=30
Modes Frame1 Frame2 Frame3 Frame4 Frame5
PSNR db
0 31.75 31.7723 31.7593 31.7743 31.755
1 31.743 31.7171 31.6864 31.6939 31.67

__________________________________________________________________________________________
The table 1 shows comparison of nine intra prediction modes
for formen cif frames 4x4 sub macroblocks with QP=30 and
table 2 shows that comparison of nine prediction modes for
mother-daughter cif frames 4x4 sub macroblocks with QP=35.
The various figs (4-22 ) are simulation results (i.e. original
frame, reconstructed frames and reconstructed frames with
deblocking filter) of mother-daughter cif frames, foreman cif
frames, news cif frames for nine intra prediction modes (0-8)
and quantization parameters-30, 35. Quantization parameter 30
for foreman cif frames, news cif frames. Quantization
parameter 30 mother-daughter cif frames.
2 31.9123 31.9029 31.878 31.891 31.8734
3 31.3911 31.3794 31.3674 31.3834 31.3826
4 31.3611 31.3585 31.3511 31.359 31.3526
5 28.6738 28.6752 28.6804 28.6671 28.6521
6 30.5477 30.5331 30.5119 30.5188 30.5123
7 27.2161 27.2159 27.2144 27.2127 27.1954
8 26.6998 26.6965 26.6952 26.692 26.6837
Compression
Ratio (%)
0 75.5339 75.5365 75.5508 75.6185 75.4362
1 75.319 75.3607 75.319 75.2826 75.2435
2 76.8424 76.8776 76.7617 76.9245 76.7943
3 68.6198 68.7096 68.8906 68.7891 68.6549
4 66.7331 66.9635 66.9193 66.9245 66.9596
5 42.625 42.6003 42.6523 42.6185 42.6042
6 67.6641 67.6849 67.6589 67.8789 67.6875
7 59.1706 59.2604 59.2995 59.2396 59.1602
8 53.1146 53.1276 53.181 53.1758 53.1393
Bit rate
0 3.5839 3.5835 3.5814 3.5715 3.5982
1 3.6154 3.6093 3.6154 3.6207 3.6264
2 3.3922 3.3870 3.4040 3.3802 3.9928
3 4.5967 4.5835 4.5570 4.5719 4.5915
4 4.8730 4.8393 4.8458 4.8450 4.8399
5 8.4054 8.4081 8.4005 8.4054 8.4075
6 4.7367 4.7336 4.7374 4.7052 4.7332
7 5.9808 5.9677 5.9619 5.9707 5.9824
8 6.8679 6.8660 6.8582 6.8590 6.8643

__________________________________________________________________________________________
Table 2

__________________________________________________________________________________________
Mode 0
Fig.4. Original frame
Fig.5. Reconstructed frame
Fig.6. Reconstructed frame with Deblocking-filter
Mode 1
Mode 2
Mode 3
Mode 4
Fig 10. Reconstructed frame
Mode 5

__________________________________________________________________________________________
Mode 6
Fig.12 Reconstructed frame
Mode 7
Mode 8
Fig.15. Reconstructed frame with Deblocking-filter
Mode 0
Mode 1
Mode 2
Mode 7

__________________________________________________________________________________________
Mode 8
Mode 0
Mode1
Mode2
Mode 7
Mode 8
CONCLUSIONS
The proposed work is selection of intra prediction modes and
best prediction mode for intra frame coding in advanced video
coding standard is carried out using Matlab. The results shows
that the reconstructed picture quality (PSNR), compression
ratio and bit rate achieved for test yuv I-frames, CIF format
(mother-daughter and foreman and news) for different
quantization parameters. Obtained results for all the nine intra
prediction modes. The mode0, mode1, mode2, mode3, mode4
and mode6 gives good PSNR, compression ratio and low bit
rate (tables I And II) compared to mode 5, mode7 and mode 8.
In the proposed method The mode0, mode1, mode2, mode3,
mode4 and mode6 gives good PSNR, compression ratio and
low bit rate, out of these modes mode 2 gives high PSNR,
compression ratio and Low Bit rate indicted yellow mark in
table(I and II), which is the best prediction mode.
REFERENCES
[1]. Iain E.Richardson,The H.264 and MPEG-4 Video
Compression :Video coding for Next-generation Multimedia,
Johan Wiley& Sons, first edition 2003.
[2]. Iain E. Richarson, The H.264 Advanced Video
Compression Standard, Johan Wiley& Sons, Second edition
2010.
[3]. Youn-Long Steve Lin, Chao-Yang Kao Hung-Chih Kuo
VLSI Design for Video Coding, Springer-2010.

__________________________________________________________________________________________
[4]. Huang Hui, Cao Tie-Yong, Zhang Xiong-wei, The
Enhanced Intra Prediction Algorithm for H.264,congress on
Image and signal processing 2008.
[5]. Chaminda Sampath Kannangara, Complexity Management
of H.264/AVC Video Compression, the Robert Gordon
University 2006.
[6]. http://www.vcodex.com

Selection of intra prediction modes for intra frame

More Related Content

What's hot (18)

Viewers also liked (20)

Similar to Selection of intra prediction modes for intra frame (20)

More from eSAT Publishing House (20)

Recently uploaded (20)

Selection of intra prediction modes for intra frame