intra frame coding for advanced video coding standard to … · 2017. 4. 7. · intra frame coding...

@IJMTER-2015, All rights Reserved 456

Intra Frame Coding for Advanced Video Coding Standard to reduce

Bitrate and obtain consistent PSNR Using Gaussian Pulse

Manjanaik.N1, Dr.Manjunath.R

2

1Electronics and Communication Engineering, Jain University, Bangalore, India.

2Senior Domain Specialist, Philips Company, Bangalore, India.

Abstract - This paper proposes Intra frame coding in Advanced Video Coding Standard to reduce bit

rate and achieve consistent PSNR using Gaussian pulse .In Intra prediction there are eight intra

prediction modes such as Mode-0 (vertical), Mode-1 (horizontal), Mode-2 (DC), Mode-3 (diagonal

down left), Mode-4 (diagonal doen right), Mode-5 (vertical right), Mode-6 (horizontal right), Mode-

7(vertical left) and Mode-8(horizontal up). Gaussian pulse is pulase that has waveform described by

Gaussian distribution. Gaussian pulse, which removes ringing, blocking artifacts and provides smooth

reconstructed image. The Gaussian pulse operation smoothens the signal. The proposed algorithm intra

frame coding implemented using matlab. In the proposed algorithm Gaussian pulse is multiplied with

each 4x4 transformed coefficients of intra frame of 4x4 sub macro block before quantization block.

The simulation results are obtained for yuv video sequences in CIF and QCIF format for different

quantization parameters with Gaussian values using Matlab. The proposed algorithm gives consistent

PSNR, and reduction in bitrate. The PSNR and bit rate achived for yuv intra frame sequences, are

tablated for different quantization parameters with Gaussian value. The

rate distortion curve, and bit rate vs quantization parameter are plotted. The simulation results shows

that the proposed algorithm can achieve controlled/consistent PSNR, and reduced in bit rate 24.4% for

CIF and 18.08% QCIF sequences.

Keywords: H.264, Gaussian pulse, Intraframe, PSNR, CAVLC, Prediction Modes.

I. INTRODUCTION H.264/AVC is the latest video coding standard jointly developed by Joint video team in 2003,which is

organized by two international standards bodies ie the Internatonal Telecommunication Union-

Telecommunications sector (ITU-T) video coding experts group and International Organization for

Standardization/ International Electro-technical Comission (ISO/IEC) moving picture experts

group.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 frame

encoded by integer transform, Gaussian pulse, quantization and context adaptive variable length coding

to get compression ratio and bit rate. In reverse process, reconstruct the picture to measure PSNR. The

existing video compression standards such as MPEG-2, MPEG-4 support adaptive changes in the

compression ratio and bitrate. The quantisation steps are varied by the appropriate choice of quantisation

matrix coefficients. The matrix may be chosen from a set of matrices defined by the standard or user can

drive altogether a new matrix in which case the coefficients need to be transmitted along with the bit

stream. A coarse step reduces the bitrate and increases compression at the cost of drastic reduction in the

PSNR of the reproduced image. so that proposed algorithm is implemented to avoid image getting bad

hit and to achive controlled PSNR and bit rate with high compression ratio in a contolled way by

multiplying Gaussian pulse each 4x4 transformed coefficients of intra frame.

International Journal of Modern Trends in Engineering and Research (IJMTER) Volume 02, Issue 06, [July– 2015]ISSN (Online):2349–9745 ; ISSN (Print):2393-8161


II. METHODOLOGY

The test yuv video sequence in QCIF and CIF format is used as an input which is open source

downloaded from the website. Read intra frames of yuv video sequence frames (Foreman QCIF anc

CIF, Suzie QCIF and tempete CIF etc). Each intra frame is processed in macroblock which is further

divided into terms of 4x4 sub blocks. Each intra frame processing is done in order to get high video

quality and high compression ratio and low bit rate as compared to previous video coding standards.

Each intra frame processing includes , 4x4 sub blocks from 16x16 macro block of intra frame, residual,

integer transformation, Gaussian pulse quantization and entropy encoding at forward path. The intra

frame is reconstructed by doing inverse process at the reconstruction path.

In the proposed algorithm each 4x4 sub macro block of intra frame is transformed into frequency

domain samples using integer transformer, these samples are multiplied with Gaussian pulse, which

smoothens the signal in reversible way to avoids picture image getting bad hit, blocking artifacts and

improves functionality of quantize block, quantized coefficients are encoded using context adaptive

variable length coding to obtain compressed bit to get compression ratio and bit rate. In the reverse

process quantized coefficients are inverse quantized and inverse transformed to reconstruct the frame,

after measuring PSNR of frame.

III. 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, Gaussian pulse, 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, transformed



coeffients are multiplied with Gaussian pulse 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].

IV. GAUSSIAN PULSE A pulse has the waveform of a Gaussian distribution, which is similar to a bell curve shape. Gaussian

pulse is shaped as a Gaussian function or Gaussian distribution.Gaussian pulse is based on Gaussian

function that is symmetric bell curve shape that quickly falls off towards zero.In the proposed method

the Gaussian pulse equation used for Intra frame coding in advanced video coding standard

G (v) =exp (-v ^ 2) ------(1)

The above equation are used for intra frame coding. After transform, frequency domain samples are

obtained, these samples are multiplied with a Gaussian pulse. The Gaussian operation smoothens the

samples. In the proposed algorithm the Gaussian pulses generated for different Gaussian scaling

parameters are shown in Figure.2-3. Matlab program written for generation of Gaussian pulses for

different Gaussian value, the generation of Gaussian pulse is considering the sampling frequency, time

base and Gaussian scaling parameter.

Fig.2. Gaussian pulses for v = 0.1, 0.2 and 0.4

Fig.3. Gaussian pulses for v = 1, 2 and 3



V.INTRA FRAME CODING

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.4 the pixels (A-M) are reconstructed previously and consider as reference pixels for

current block. The pixels labeled (a-m) are prediction pixels.

Fg. 4. 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.5 shows the intra prediction modes.

Fig.5. 4x4 intra prediction modes

In H.264/AVC standard, spatial redundancy is removed in intra frame by using nine intra prediction

modes. All nine modes are used for encoding intra 4X4 luma block of intra frame, modes 0, 1, 3, 4, 5, 6,

7, 8 are directional prediction modes and mode 2 is the DC prediction mode with no direction. Mode 1

specifies the vertical prediction mode,for mode 0 prediction samples formed by subtracting

reconstructed samples (A B C D) with current block samples. Mode 1 specifies the horizontal prediction

mode, for mode 1 prediction samples formed by subtracting reconstructed samples (I J K L) with current

block samples. Mode 1 specifies DC mode prediction for 4X4 block is to replace all pixels/samples in

the current 4X4 block by the mean value of the reconstructed samples (A B C D I J K L). for modes 3 to

8 prediction samples formed from a weighted average of the prediction samples A-M.

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



VI. OPERATION OF DC PREDICTION MODE

There are nine prediction modes in intra prediction for Intra frame coding in which one of prediction

mode steps involved are explained.the steps performed in DC mode for Intra frame coding as follows.

For DC prediction mode, all the pixels of current sub block is predicted by mean value of upper and left

reconstructed pixels of reconstructed block. Prediction equation generated for DC mode as follows

(A+B+C+D+I+J+K+L)/8 = S ----(2)

a1= a2 = a3 = a4 = a5 = a6 = a7 = a8 = a9 = a10 = a11 = a12 = a13 = a14 = a15 = a16 = S

The operation involved in DC prediction mode for intra frame is as shown in Figure 6.

Fig 6. DC intra prediction mode for intra frame

VII. STEPS OF IMPLEMENTATION

The following steps are to required for intra frame coding in AVC using Gaussian pulse.

• Input test yuv video sequence in QCIF and CIF format is used as an input .

• Read Intra frame from input test yuv video sequence (I-frame).

• Each frame is processed in terms of macro block; macro block consists of 16x16 pixels.

• A 16x16 macro block is further divided into a 4x4 sub macro block.

• A first 4x4 sub block is processed directly without using previously reconstructed block followed by integer transform, Gaussian pulse, quantization, entropy encoding at encoder and reverse process

at reconsrction path of encoder.

• Reconstruct a 4x4 sub block using inverse process with DC intra prediction mode

• Obtain residual block by subtracting current sub block with previously reconstructed sub block.

• Residual of 4x4 sub block is integer transformed, Gaussian pulse, quantized and entropy encoded at encoder and reverse process such as inverse quantized, inverse transform and add prediction at

reconstruction path of encoder.

• Finally measure PSNR, bit rate and compression ratio for different QP with Gaussian value.



VIII. IMPLEMENTAION

The proposed work is carried out using Matlab. The input is test yuv video sequences in QCIF and CIF

format. A Matlab program is written, which reads the test yuv video sequence and extracts only Intra

frame. The next process involves intra frame is divided Macro block into 4x4sub block which 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. Obtain the residual block by subtracting reconstructed bock with

current block. Residual coefficients are integer transformed, transformed coefficients (i.e. frequency

samples) are multiplied by Gaussian pulse, these coefficients are, quantized and encoded using context

adaptive variable length coder (CAVLC) to get compressed bit. At H.264 encoder in the reconstruction

path, perform reverse processes to get reconstruct image and finally measure quality picture (PSNR) for

Mode-2 DC prediction mode.

Measure PSNR, compression ratio and bit rate for different quantization parameters with Gaussian

value.

IX. RESULTS AND DISCUSSION

The proposed work Intra frame coding is done using Gaussian pulse. The results obtained for the yuv

video sequences in CIF and QCIF frames and different quantization parameters and Gaussian value for

Mode-2 4x4 intra DC prediction mode.

Table I Results of Foreman QCIF Hung et al [2008] algorithm

QP

PSNR (dB) Bitrate (kbits/s)

JM 8.6

(H.264)

Proposed JM 8.6

(H.264)

Proposed

24 39.92 39.95 883.72 862.48

28 37.12 37.12 619.30 598.61

32 34.30 34.29 413.40 398.21

36 31.60 31.58 274.70 263.09

Remarks PSNR 0.01 ∆Bitrate = -3.42 %

Table II Results of Foreman QCIF Proposed algorithm

QP

PSNR (dB) Bitrate (kbits/s)

JM18.

6

(H.26

4)

Proposed JM18.6

(H.264

)

Propos

ed

24 37.00 37.58 424.32 254.55

28 36.82 37.63 312.96 194.86

32 34.07 37.63 218.40 147.69

36 31.89 37.72 148.32 133.34

Remark

s

PSNR consistent ∆Bitrate = -30.04

%



Table III Results of Suzie QCIF Hung et al [2008] algorithm

QP PSNR (dB) Bitrate (kbits/s)

JM 8.6

(H.264)

Proposed JM 8.6

(H.264)

Proposed

24 40.59 40.62 700.97 691.51

28 37.80 37.79 464.08 457.59

32 35.20 35.21 288.23 282.11

36 32.94 32.93 176.79 173.70


Table IV Results of Suzie QCIF Proposed algorithm


JM18.6

(H.264)

Proposed JM18.6

(H.264)

Proposed

24 39.85 37.31 306.16 240.24

28 37.28 37.35 208.48 180.84

32 34.94 37.38 134.36 131.75

36 32.79 37.50 86.24 96.80

Remarks PSNR consistent ∆Bitrate = -6.12 %

Table V Results of Foreman CIF Hung et al [2008] algorithm


JM 8.6

(H.264)

Proposed JM 8.6

(H.264)

Proposed

24 40.25 40.27 2699.38 2659.59

28 37.77 37.76 1777.41 1746.19

32 35.18 35.18 1121.13 1102.18

36 32.87 32.86 703.62 694.30


Table VI Results of Foreman CIF Proposed algorithm


JM

18.6

(H.264)

Proposed JM18.6

(H.264)

Proposed

24 37.00 37.58 424.32 254.55

28 36.82 37.63 312.96 194.86

32 34.07 37.63 218.40 147.69

36 31.89 37.72 148.32 133.34

Remarks PSNR consistent ∆Bitrate = -

18.62%



Table VII Results of Tempete CIF Hung et al [2008] algorithm


JM 8.6

(H.264)

Propos

ed

JM

8.6

(H.26

4)

Propose

d

24 39.46 39.49 5547.

19

5502.9

3

28 36.11 36.12 4116.

46

4087.8

0

32 32.73 32.74 2843.

07

2811.5

8

36 29.63 29.63 1849.

85

1835.5

6

Remar

ks

PSNR 0.02 ∆Bitrate = -1.24

% Table VIII Results of Tempete CIF Proposed algorithm


JM18.6

(H.264)

Propos

ed

JM18.6

(H.264)

Propose

d

24 38.67 35.35 2370.3

2

1363.0

5

28 35.32 35.37 1743.7

6

1104.7

2

32 31.94 35.41 1217.2

8

877.35

36 28.98 35.48 805.20 694.70

Remar

ks

PSNR consistent ∆Bitrate = -

18.62%

The results obtained for the test yuv sequences such as foreman, tempete and suzie in CIF and QCIF

format, for quantization parameter (24,28,32 and 36) with gaussian value(0.1).The table I, II,

III,IV,V,VI,VII and VIII gives PSNR, and bit rate of intra frame of yuv CIF frames for 4x4 intra DC

prediction mode with different quantization parameters and Gaussian Value 0.1. The rate distortion

curve, bitrate curve and compression ratio curve for different quantization parameters with scaling factor

v=0.1 of DC prediction mode is shown in figure 7-10.



Table IX Performance comparison proposed method with previous work Sequence

QCIF

Method Parameters QP Bit rate reduction

Proposed

Bit rate

Reduction

Hung et al [2008]

Inference

24 28 30 36

Foreman

QCIF

JM 18.6 PSNR (dB) 37.00 36.82 34.07 31.89

-30.04 %

-3.42 %

Bit rate reduction

achieved more

Compared

With

Hung et al [2008]

PSNR is consistent

Bit rate (kbps) 424.32 312.96 218.40 218.40

Proposed PSNR (dB) 37.58 37.63 37.63 37.72

Bit rate (kbps) 254.55 194.86 147.69 133.34

Suzie

QCIF

JM 18.6 PSNR (dB) 39.85 37.28 34.94 32.79

-6.12 %

-2.11 % Bit rate (kbps) 306.16 208.48 134.36 86.24

Proposed PSNR (dB) 37.31 37.35 37.38 37.50

Bit rate (kbps) 240.24 180.84 131.75 96.80

Average JM 18.6, Intra frame, frame rate 30 frames per second CAVLC

-18.08 % -2.765%

Foreman

CIF

JM 18.6 PSNR (dB) 39.88 37.46 35.04 32.69

-18.62%

-1.75 % Bit rate (kbps) 1256.08 870.64 585.60 389.44

Proposed PSNR (dB) 38.98 38.98 39.03 39.15

Bit rate (kbps) 914.77 673.72 487.64 358.30

Tempete CIF JM 18.6 PSNR (dB) 38.67 35.32 31.94 28.98

-30.18%

-1.24 % Bit rate (kbps) 2370.32 1743.76 1217.28 805.20

Proposed PSNR (dB) 35.35 35.37 35.41 35.58

Bit rate (kbps) 1363.05 1104.72 877.35 694.70

Average JM 18.6, Intra frame, frame rate 30 frames per second CAVLC

- sign indicates saving

+ sign indicates increasing

-24.4 %

-1.49%

The graph of rate distortion for different quantization parameters with scaling factor v = 0.1 indicates

that for different quantization,consistent PSNR and reduced in bit rate achieved. The bit rate for

different quantization parameters Gaussian value v = 0.1 is indicates that for different quantization,

faster reduced in bit rate achieved compared with JM reference 18.6 software of H.264 standard.

Figure.11-12 indicates reconstructed frames Foreman and tempete in CIF format for different

quantization paremeter with Gaussian scaling factor v=0.1.

The performance parameters are measured using the following formula.

∆P = PSNR(Proposed) – PSNR(JM reference) ---(3)

∆B = Bit rate(proposed)- Bit rate(JM reference) X 100 ---(4)

Bit rate(JM reference)

Fig.7. Bitrate vs Quantization parameter

20 25 30 35 400

50

100

150

200

250

300

350

400

450

500

550Bitrate vs QP Foreman Qcif

Bitra

te (

kbps

)

Quantization Parameter (QP)

Bitrate Foreman Qcif Proposed

Bitrate Foreman Qcif JM 18.6






Table I,III, V and VII indicates the performance parameters of Hung et al [2008] algorithm.

Table II,IV, VI and VIII indicates the performance parameters of proposed algorithm.

The table IX gives over all performance comparision of proposed method and previous method.From the

table it observe that bit rate reduction achieved more about 24.4% for CIF sequence and 18.08% for

20 25 30 35 400

50

100

150

200

250

300

350

400Bitrate vs QP Akiyo Qcif

Bit

rate

(k

bp

s)


Bitrate Akiyo Qcif Proposed Method

Bitrate Akiyo Qcif JM 18.6

23 24 25 26 27 28 29 30 31

800

1000

1200

1400

1600

1800

2000

2200

2400

Bitrate vs QP tempete cif

Bitra

te (

kbps)


Bitrate tempete cif Proposed Method

Bitrate tempete cif JM 18.6

0 50 100 150 200 250 300 350 400 450 500 55020

25

30

35

40

45

50Rate Distortion Foreman qcif

Peak S

ignal to

Nois

e R

atio (

dB

)

Bit rate(kbps)

RD curve Foreman qcif Proposed Method

RD curve Foreman qcif JM18.6



QCIF sequence compared with previous algorithm.PSNR is consistent in the proposed method but

PSNR is not consistent in the prevous algorithm.

From the ratedistortion curve is observe that PSNR is consistent. From the bit rate curve it observe that

bit rate reduces faster in proposed algorithm.

Fig.11.Reconstucted Foreman Intra frame cif

Fig.12. Reconstrcucted tempete frame cif

X. CONCLUSIONS The proposed work is Intra frame coding in AVC using Gaussian pulse for 4x4 intra DC prediction

mode. It is observe that using Gaussian pulse can achieved consistent PSNR and averagely 24.4%

reduced in bit rate for CIF sequence and 18.08% for QCIF sequence.The simulated results are presented

in table (I-VIII) and rate distortion curve, and bit rate curve are represented. The results shows that

consistent reconstructed picture quality (PSNR), reduced in bit rate achieved.

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] Chaminda Sampath Kannangara, Complexity Management of H.264/AVC Video Compression, the Robert Gordon

University 2006.

[4] Thomas Wiegand, Gory. Sullivan, Seniour Member, IEEE, Gisle Bjontegaard and Ajay Luthra, “Overview of the

H.264/AVC Video Coding Standard”, IEEE Transactions on circuits and systems for video Technology, Vol. No 7 2003.

[5] Rein van den Boomgaard and Rik van der Weij, “Gaussian Convolutions Numerical Approximations Based on

Interpolation”, Intelligent Sensory Information Systems, University of Amsterdam, and The Netherlands. [6] Pascal Gwosdek, Sven Grewenig1, Andr´es Bruhn, and Joachim Weickert, “Theoretical Foundations of Gaussian

Convolution by Extended Box Filtering”.

[7] Huang Hui, Cao Tie-Yong,Zhang Xiong-Wei, “The enhanced intra prediction algorithm for H.264”IEEE 978-0-7695-

3119-9/2008 Congress on Image and Signal Processing.

[8] Jijun Shi, Yunhui Shi, Baocai Yi, “ Fast intra prediction mode decision for AVC based on edge direction detection” 2008

The Institution of Engineering and technology.

[9] K.Bharanitharam, An-Chao Tsai, “Efficient Block Size Decision Algorithmm for intra mode decision in AVC encoder”

2009 IEEE International Symposium on Multimedia

[10] http://www.vcodex.com

intra frame coding for advanced video coding standard to … · 2017. 4. 7. · intra frame coding...

Documents