[ieee international conference on advanced nanomaterials & emerging engineering technologies...

6
I Proceedings of the "International Conference on Advanced Nanomaterials & Emerging Engineering Technologies" (ICANET-2013) L . _· . organtzed by Sathyabama University, Chennai, India in association with DRDO, New Delhi, India, 24t h _26t h , July, 2013. Entropy Coding ofH.264/AVC using Exp-Golomb Coding and CAVLC Coding Savita Nargundmath#l, Archana Nandibewool2 # CSE department SDMCET Dhala Kamataka, India. '[email protected] '[email protected] Abstract- As the costs for both processing power and memory have reduced, network support for coded video data has diversified, and advances in video coding technology have progressed, the need has arisen for an industry standard for compressed video representation with substantially increased coding efficiency and enhanced robustness to network environments. The H.264/AVC standard aims to enable significantly improved compression performance compared to all existing video coding standards. In this project we employ two schemes of coding transform coefficients namely exponential Golomb coding & context adaptive variable length eoding (CA VLC).And the major part of the contribution is the decoding strategy applied for decoding which results in performance enhancement saving the memory and decoding time which are the most important factors for bandwidth utilization. The transform coeffieients are obtained using a simple zigzag scan technique. The consensus among the major players of the eommunications and video industry on H.264 might provide the major thrust for this new standard. I. INTRODUCTION Video compression plays an important role in enabling video processing on multimedia systems. Video compression is essential in order to reduce the storage requirements to manageable levels and to transmit data with the existing channel capacities. The video codec that achieves the highest data compression without sacrificing on the picture quality is the MPEG-4 Part 10 Advanced Video Coding. The Joint Video Team (JVT) of ISO/IEC MPEG and ITU-T VCEG are finalising a new standard for the coding (compression) of natural video images. The new standard will be known as H.264 and also MPEG-4 Part 10, "Advanced Video Coding". The standard specifies various types of entropy coding: Context adaptive variable length Coding (CA VLC), Context-based Adaptive Binary Arithmetic Coding (CABAC) and Exp-Golomb coding. Thus, an efficient implementation of H.264/AVC codec should be developed. Recently, the implementation of the H.264/AVC decoder has been studied by some researchers Jun-Young Lee, Jae-Jin Lee, and Seong Mo Park analysed the complexities involved in decoding in the H.264/AVC using New Lookup Tables and proposed an optimized implementation of designing a codec which is used for Entropy coding but One critical disadvantage of this method is the unnecessary iterations even for decoding one syntax element. And also In order to solve the drawbacks of the conventional method, several methods were proposed in the literature [1] so here in this paper we are trying to decode it efficiently overcoming the drawback of all the previous methods specified. The proposed methods in the paper can be adopted for all the variants of CAVLC decoding. And we are going to implement both the Exp-Golomb coding and CAVLC Coding where we are referring the literature [2]- [5] and finally making the Performance analysis of the above two techniques. In this part we will be measuring and comparing some of the important parameters of the above two coding strategies. The parameters are coding and decoding time taken and the bit rate of both the methodologies. These parameters are to differ with different input transform coefficients. This paper is organized as follows: Introduction to the entropy coding is presented in section I of the paper. The algorithms for processing of Exp-Golomb and CAVLC processors is presented in Section 1I[2].This is followed by the conventional method of decoding and its drawback in the next section[l].Our proposed method of decoding and proving it is as effecient in section IV. The comparison of both the Golomb and CAVLC coding techniques in terms of bitrate and time taken for encoding and decoding is presented in section V.While the results are discussed in Section V. The conclusion is presented in the last section. II. ALGORITHM FOR PROCESSING EXP-GOLOMB CODING AND CONTEXT ADAPTIVE VARIABLE LENGTH CODING. A H.264 video encoder consists of various functional modules such as an intra-prediction processor, an integer transform and quantization processor and a CAVLC processor. Intra-prediction is done in order to exploit the spatial redundancy within a frame of a motion picture. Each pixel is predicted based on the values of its neighbouring pixels that are available. A video sequence is input to the integer transform and quantization processor. Vectoring 2-D matrix of quantized DCT coefficients are represented in the form of a single-dimensional vector. After quantization, most of the high frequency coefficients (lower right corner) are zero. The quantized coefficients are then fed to the Golomb or CAVLC module, which assigns variable length codes to get the desired compressed bit stream. The quantized coefficients are a 4x4 block of data which is scanned in a zig-zag manner and to exploit the number of zeros, a zig-zag scan of the matrix is used. Zig-zag scan allows all the DC coefficients and lower frequency AC coefficients to be scanned first. The DC coefficients are encoded using differential encoding and AC coefficients are encoded using run-length encoding and then a 978-1-4799-1379-4/13/$31.002013 IEEE 607

Upload: archana

Post on 05-Apr-2017

212 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: [IEEE International Conference on Advanced Nanomaterials & Emerging Engineering Technologies (ICANMEET-2013) - Chennai (2013.7.24-2013.7.26)] International Conference on Advanced Nanomaterials

I� Proceedings of the "International Conference on Advanced Nanomaterials & Emerging Engineering Technologies" (ICANMEET-2013)

L.�""_·L:I . ...... organtzed by Sathyabama University, Chennai, India in association with DRDO, New Delhi, India, 24th _26th, July, 2013.

Entropy Coding ofH.264/AVC using Exp-Golomb Coding and CA VLC Coding

Savita Nargundmath#l, Archana Nandibewool2 # CSE department SDMCET Dhanllad, Kamataka, India.

'[email protected] '[email protected]

Abstract- As the costs for both processing power and

memory have reduced, network support for coded video data has diversified, and advances in video coding technology have

progressed, the need has arisen for an industry standard for

compressed video representation with substantially increased

coding efficiency and enhanced robustness to network

environments. The H.264/AVC standard aims to enable

significantly improved compression performance compared to all

existing video coding standards. In this project we employ two

schemes of coding transform coefficients namely exponential

Golomb coding & context adaptive variable length eoding

(CA VLC).And the major part of the contribution is the decoding

strategy applied for decoding which results in performance

enhancement saving the memory and decoding time which are

the most important factors for bandwidth utilization. The

transform coeffieients are obtained using a simple zigzag scan

technique. The consensus among the major players of the

eommunications and video industry on H.264 might provide the

major thrust for this new standard.

I. INTRODUCTION

Video compression plays an important role in enabling video processing on multimedia systems. Video compression is essential in order to reduce the storage requirements to manageable levels and to transmit data with the existing channel capacities. The video codec that achieves the highest data compression without sacrificing on the picture quality is the MPEG-4 Part 10 Advanced Video Coding.

The Joint Video Team (JVT) of ISO/IEC MPEG and ITU-T VCEG are finalising a new standard for the coding (compression) of natural video images. The new standard will be known as H.264 and also MPEG-4 Part 10, "Advanced

Video Coding". The standard specifies various types of entropy coding: Context adaptive variable length Coding (CA VLC), Context-based Adaptive Binary Arithmetic Coding (CABAC) and Exp-Golomb coding. Thus, an efficient

implementation of H.264/A VC codec should be developed. Recently, the implementation of the H.264/A VC decoder has been studied by some researchers Jun-Young Lee, Jae-Jin Lee, and Seong Mo Park analysed the complexities involved in

decoding in the H.264/A VC using New Lookup Tables and proposed an optimized implementation of designing a codec which is used for Entropy coding but One critical disadvantage of this method is the unnecessary iterations even

for decoding one syntax element. And also In order to solve the drawbacks of the conventional method, several methods were proposed in the literature [1] so here in this paper we are trying to decode it efficiently overcoming the drawback of all

the previous methods specified. The proposed methods in the

paper can be adopted for all the variants of CA VLC decoding.

And we are going to implement both the Exp-Golomb coding and CA VLC Coding where we are referring the literature [2]­[5] and finally making the Performance analysis of the above

two techniques. In this part we will be measuring and comparing some of the important parameters of the above two coding strategies. The parameters are coding and decoding time taken and the bit rate of both the methodologies. These

parameters are to differ with different input transform coeffi ci ents.

This paper is organized as follows: Introduction to the entropy coding is presented in section I of the paper. The

algorithms for processing of Exp-Golomb and CA VLC processors is presented in Section 1I[2].This is followed by the conventional method of decoding and its drawback in the next section[l].Our proposed method of decoding and proving it is

as effecient in section IV. The comparison of both the Golomb and CA VLC coding techniques in terms of bitrate and time taken for encoding and decoding is presented in section V.While the results are discussed in Section V. The

conclusion is presented in the last section.

II. ALGORITHM FOR PROCESSING EXP-GOLOMB CODING AND CONTEXT ADAPTIVE VARIABLE LENGTH CODING.

A H.264 video encoder consists of various functional modules such as an intra-prediction processor, an integer transform and quantization processor and a CA VLC processor. Intra-prediction is done in order to exploit the spatial redundancy within a frame of a motion picture. Each pixel is predicted based on the values of its neighbouring pixels that are available. A video sequence is input to the integer

transform and quantization processor. Vectoring 2-D matrix of quantized DCT coefficients are represented in the form of a single-dimensional vector. After quantization, most of the high frequency coefficients (lower right corner) are zero. The

quantized coefficients are then fed to the Golomb or CA VLC module, which assigns variable length codes to get the desired compressed bit stream. The quantized coefficients are a 4x4 block of data which is scanned in a zig-zag manner and to exploit the number of zeros, a zig-zag scan of the matrix is used. Zig-zag scan allows all the DC coefficients and lower frequency AC coefficients to be scanned first. The DC coefficients are encoded using differential encoding and AC

coefficients are encoded using run-length encoding and then a

978-1-4799-1379-4/13/$31.002013 IEEE 607

Page 2: [IEEE International Conference on Advanced Nanomaterials & Emerging Engineering Technologies (ICANMEET-2013) - Chennai (2013.7.24-2013.7.26)] International Conference on Advanced Nanomaterials

I� Proceedings of the "International Conference on Advanced Nanomaterials & Emerging Engineering Technologies" (ICANMEET-2013)

L.�"""'_L:l. _ organized by Sathyabama University, Chennai, India in association with DRDO, New Delhi, India, 24th _26th, July, 2013.

4x4 block of data which is scanned in a zig-zag manner is then fed to the Golomb or CA VLC processor.

The algorithms for processing the Golomb coder is shown in "Fig. 1 and 2".

a>O Input >-__ a<_o----, am

a=O

L-__ I rod!um �I ____ �

ENCODER Fig! :Flowchart for Exp-Golomb Encoder

/ lnPut COd

Y word

Compute length & decimal value of codeword

Find M and info rl Calculate � <0

L-_c __ ode

-,-num

c:-----__ >0

1-0 I a=-[codenuml2] I ar I I a=[(codenum+ l )I2] 1

[ Done I

DECODER Fig2:Flowchart for Exp-Golomb Decoder

A. Exp-Golomb coder:

As we can see the algorithm for both Golomb encoder and

decoder in the above figures we are interested in the generation of the desired codeword from the given code _ num value which is done in the Golomb encoder and also we will be able to extract the code_num value by the codeword which

is generated by the encoder with the help of Golomb Decoder. The complete steps for it are specified in literature [2].Below is the description for the algorithm flow:

Exp-Golomb codes (Exponential Golomb codes) are variable length codes with a regular construction. the codewords progress in a logical order. Each codeword is constructed as follows:

[M zeros][lHINFO] where INFO is an M-bit field carrying information.

The first codeword has no leading zero or trailing INFO; codewords 1 and 2 have a single-bit INFO field; codewords 3-6 have a 2-bit INFO field; and so on. The length of each codeword is (2M+l) bits. Each Exp-Golomb codeword can be constructed by the encoder based on its index code_Dum:

Me ?log2(code _num+ I)? INFO = code num + 1- 2 /\ M

A codeword can be decoded as follows:

l. Read in M leading zeros followed by l. 2. Read M-bit INFO field. 3. code num = 2M + INFO - 1

(For codeword 0, INFO and M are zero).

The algorithm for processing the CA VLC coder is as shown in Fig.3

Is num of JJ�.?

Isrunbeforetab� matched .. ?

Is runbeforeoonzertl coelFlJ .. ?

Figure 3: Flowchart for processing CA VLC

B. CA VLC Encoder:

This algorithm forms the main part of the H.264 compression coding standard. Its Context Adaptive, that is the bit stream generated changes with the input 4X4 matrix of transform coefficients. The context adaptive part is the most advantageous aspect of this algorithm.

But similarly the complexity of implementation is greater.

There are various ITU-T recommended tables, algorithms and rules to be followed while encoding of the bit stream. All

608

Page 3: [IEEE International Conference on Advanced Nanomaterials & Emerging Engineering Technologies (ICANMEET-2013) - Chennai (2013.7.24-2013.7.26)] International Conference on Advanced Nanomaterials

I� Proceedings of the "International Conference on Advanced Nanomaterials & Emerging Engineering Technologies" (ICANMEET-2013)

L.�"""_L.:I . ..... organized by Sathyabama University, Chennai, India in association with DRDO, New Delhi, India, 24th _26th, July, 2013.

these tables are prescribed by the ITU-T. The CAVLC encoder is designed such that the encoder by itself will be able to extract those values from those tables to obtain the required coded bitstream.

Again, the inverse operation of converting the bit stream back to the 4X4 transform matrix is done by the decoder. Here too the decoder by itself will be able to obtain the final resultant matrix using the ITU-T tables without any complexities involved in it. The literature paper [2][5] gives the complete designing methodology for CA VLC encoding and decoding.

CA VLC encoding and decoding of a block proceeds as

follows. The complete flow of algorithm goes with following steps:

1) coejJ token: Compute the median value (nC) of the number of nonzero coefficients in upper and left blocks if

exists. The median value determines the VLD table number for current coejJ token decoding. After deciding which table to use, coejJ token syntax element encoding both the total number of nonzero coefficient (TotaICoejJ ) and the number

of trailing +1/-1 values (TrailingOnes) is ready to be parsed.

There are one fixed-length code table and three variable­length code tables for luminance and one variable-length code table for chrominance.

2) TrailingOnes: For each TrailingOnes (+11 - 1)

signaled by coejJ token, the sign is decoded with a single bit 0 representing the positive and 1 representing the negative value. Simple arithmetic operations can be used to decode it.

3) level: Decode level coefficients using one of seven VLD tables if TotalCoejJ and TrailingOnes do not have the same number. The selection of the table is based on the

magnitude of the most recently decoded coefficients. The decoding sequence can further be partitioned into four phases: 1) decode level prefix; 2) generate length; 3) generate LevelSuffix; and 4) generate level.

4) total zeros: Decode the number of all zeros preceding the highest frequency nonzero coefficient if the Total-CoejJ is greater than O. The 15 different variable-length tables are considered depending on previously decoded TotalCoejJ.

5) run before: Decode the number of zeros preceding each nonzero coefficient starting with the highest frequency nonzero coefficient working back toward the DC coefficient using run before VLD tables. Unlike previous syntax

elements, multiple run before decoding is executed until there are no more zeros left to decode or only one nonzero coefficient left. The VLD tables of run before syntax element consist of seven tables that other hand TrailingOnes and level

are decoded by simple arithmetic operations since these two syntax elements are encoded by regular VLC codes.

III. CONVENTIONAL METHOD OF DECODING.

Each codeword in each VLD(Variable length decoding) table can be represented by a unique pair of length and

codevalue.These VLD tables are standard tables specified by ITU-T Recommendation refer literature[6] for the VLD tables. For example, for VLD table 0 of the coejJ token syntax element, codeword "0000100" is represented by (7,4) where

the first digit represents the length and the second digit represents code value. Accordingly, two separate tables with length table ("Ientab") and code value table ("codtab")are constructed as depicted. The index pair of (3, 5) matched to the length (=7) and code value (=4) at the same time is the desired TrailingOnesand TotaICoeff.respectively.Observe the below specified conventional decoder where The "Shifting

BujJer (SB) " outputs the codevalue of "code value bit" having length of "length" from the "Input BujJer" with big-endian fashion. Two code values from both "SB" and "Table LookUp

(TLP) " are then compared in the"Compare and Select (CS)" in the Fig. 4. The iterations are repeated until the two code values are same.

(l,y)

Table Lookllp (TLP)

Initial value (x,)')" (0,0)

('0

Update or allay

Codevllltl( ,ab

Compar� alld

Seletl (CS)

DECODED YM80LS

Fig. 4:Decoding method of the conventional method Let us consider,Length of overall string= 2I,coded value

length=4,total number of codewords generated is 21x4, 14 null values(=O) Since each coejJ token VLD table has 2Ix4

codewords and fourteen null value (=0) in each table and in this conventional method right to left scan is applied in order to find the desired value so here in this case required number of iterations will be 2I-iterations for finding TotalCoejJ and TrailingOnes which is the desired value need to be find but it is happen in the worst case. Furthermore, the codewords in the table are constructed with no consideration of symbol probabilities. To illustrate the effects mentioned above on the CA VLC decoding performance, Fig. 5 shows the CA VLC decoding time analysis performed with JM reference software We notice that the decoding time of coejJ token, total zeros, and run before is more than 80% of the total decoding time.

Therefore we focus on designing the efficient decoding method of coejJ token, total =eros and run before syntax elements rather than TrailingOnes and level syntax elements. In order to solve the drawbacks of this conventional method, several methods were proposed were proposed earlier which is given in the literature paper [1 ].

609

Page 4: [IEEE International Conference on Advanced Nanomaterials & Emerging Engineering Technologies (ICANMEET-2013) - Chennai (2013.7.24-2013.7.26)] International Conference on Advanced Nanomaterials

I� Proceedings of the "International Conference on Advanced Nanomaterials & Emerging Engineering Technologies" (ICANMEET-2013)

L.�""""'IIL:I' .... organized by Sathyabama University, Chennai, India in association with DRDO, New Delhi, India, 24th _26th, July, 2013.

run_befQ� (28.6%)

Fig. 5. CA VLC decoding time (%) analysis of Foreman test sequence

IV. PROPOSED METHOD

So here we are proposing one such method which is proved to more efficient to the existing ones. Let us consider, Length of overall string= 2l,coded value length=4,total number of

codewords generated is 2lx4,14 null values(=O) One critical disadvantage of the conventional method is the unnecessary iterations even for decoding one syntax element [i.e., the iteration number n in (1) is large].This disadvantage is mainly due to scan method which was applied that is right to left scan, Because in this type of decoding method we are scanning the bit stream from right to left but our desired value will be in the first half of the bitstream, which will be on left hand side of

bitstream and scanning along the right direction would result in more iterations and also degrading the performance in terms of time and complexity. Let us consider one example ,referring to the previously defined case of desired

value being 7 and we have the code word for 7 that is 0000 0111 so here in our proposed method we are going to make the change in scan method .so here instead of scanning the given bitstream from right to left will scan the bitstream from left to right.This left to right scan will conversely going to reduce the number of iterations required.As we apply left to right scan our desired value is in the first half of the bitstream that is towards left and hence it just needs 7 iterations to fetch the desired value in order to get the desired coded value. Applying this left to right scan to the above stated example=2l,coded length value=4 bits so start scanning from left to right we obtain our desired value jus for the th iteration

so eventually this proves that extra unnecessary(2l-7=14) iterations are saved and conversely results in the performance enhancement.Here is the formula the performance analysis for obtaining the desired value,

P=(total bit stream length)-(no of iterations in which we obtained the desired value)/(total bit stream length).As per above example performance analysis is given in the below specified equation,

i.e P=( 2l-7)/2lxlOO=66.6% So the inference from this resultant performance value of 66.6% obtained results in saving the time and also saving the memory to store the value

which will eventually reduce the complexity and leads In performance enhancement.

V. THE COMPARISION OF BOTH EXP-GOLOMB AND CA VLC CODING TECHNIQUES.

In this we are making the Performance analysis of the above two techniques that is Exp-Golomb and CA VLC. In

this part we will be measuring and comparing some of the important parameters of the above two coding strategies. The parameters are coding and decoding time taken and the bit rate of both the methodologies.

Comparing the resultant bitstream generated by the final decoder of both Golomb and CA VLC shows the number of

bits reduced in both of the decoders. As we all know lesser is

the length of the bitstream the more compression is achieved.

So the comparison results showed that CA VLC has more bit reduction then to the Golomb.The context adaptive part is the most advantageous aspect of this algorithm. Which makes it possible to achieve more compression without any data 10ss.But similarly the complexity of implementation is greater. Observe down the below graph which clearly predicts the performance enhancement when used a CA VLC decoding scheme applied with our decoding strategy there is reduction

in bits in CAVLC compared to that of Golomb coding.

" · " · · " " " " , 0

_ Golomb

_ CAVLC

H.264 Graphical User Interface

� _CAV\.C GoIornbCoding

I. CAVLCCoding

Fig 6: comparison graph obtained as a result of comparison of both coding schemes in terms of bit reduction.

3 4 Type of Coding 5<:home

= El

Fig 7: Comparison graph obtained as a result of comparison of both coding schemes in terms of CPU execution time.

610

Page 5: [IEEE International Conference on Advanced Nanomaterials & Emerging Engineering Technologies (ICANMEET-2013) - Chennai (2013.7.24-2013.7.26)] International Conference on Advanced Nanomaterials

I� Proceedings of the "International Conference on Advanced Nanomaterials & Emerging Engineering Technologies" (ICANMEET-2013)

L.�"""_L.:I . ..... organized by Sathyabama University, Chennai, India in association with DRDO, New Delhi, India, 24th _26th, July, 2013.

V. CONCLUSION

In H.264 CA VLC decoding, coeff token, total zeros, and run before decoding are based on VLD tables, which are the most critical parts of whole CA VLC decoding procedures. There were several works to be able to decode three syntax elements memory-efficiently and rapidly. Various methods were evaluated based on our dual definitions of the iteration bound (i) and P=performance. The fact that the previous works could not satisfy the two design parameters simultaneously makes the efficient implementation difficult.

Therefore we propose the simpler but faster CA VLC decoding method for these three syntax elements. In this paper, new codeword structure tables and searching algorithms for fast

and efficient coeff token,total zeros and run before were developed. The proposed methods can decode the CA VLC decoding is faster than the conventional one with less complexity as the number of iterations are reduced while decoding which eventually enhances the performance and saves the time required to decode the desired value from the lengthier bitstream.

f'f. » And finally Comparing both the Golomb and CA VLC

coding techniques.The parameters are coding and decoding

time taken and the bit rate of both the methodologies which are been compared. So comparing the resultant bitstream generated by the final decoder of both Golomb and CA VLC the CA VLC has more bit reduction then to the Golomb which

is shown in the above graph Fig 6.And for the aspect of time taken we make use of different test cases for comparison. Testing is being performed for the different variants of the elements in the given quantized input matrix .Different

variants like the Trailling ones that is Tis conditions number of levels in the given input matrix or the presence or absence of the DC coefficients are the critical conditions which determine the results of the Encoders and Decoders.

2)Exp-Golomb Decoder:

Taking in consideration with few of the Test cases we can obseverve figure 7. with its expected results we are indulged in knowing the CPU execution time which eventually help us in proving that the CPU execution time is more for CA VLC

then compared to Golomb because of the complexity of the Algorithm. But we can also say it may vary sometimes because CA VLC coding scheme itself determines that it is context adaptive in nature. Thus the two parameters of bits reduction and time taken is being analyzed for the performance analysis of the both Golomb and CA VLC compression techniques.

The context adaptive part is the most advantageous aspect

of this algorithm. But similarly the complexity of implementation is greater.

VII. RESUL TS:

A . Exp-Golomb coding Results:

l)Exp-Golomb Encoder:

611

A

,® NewIO�lLAB?Walchthis�seeQlmz o.,e.d �

k ?

-, o o

-, , o o

B. CA VLC coding Results:

1) CAVLC Encoder :

'j

Page 6: [IEEE International Conference on Advanced Nanomaterials & Emerging Engineering Technologies (ICANMEET-2013) - Chennai (2013.7.24-2013.7.26)] International Conference on Advanced Nanomaterials

I� Proceedings of the "International Conference on Advanced Nanomaterials & Emerging Engineering Technologies" (ICANMEET-2013)

L.� .... __ ._", organized by Sathyabama University, Chennai, India in association with DRDO, New Delhi, India, 24th _26th, July, 2013.

[Q) New to MATlAB?Walchthis� S�� Of rr:ad Gfiting Starttd. Enter the 4"1 matrix input-[O :3 -1 0

0 -110 1000 000 OJ

-1 -1

Column:!! 1 through Ii

Column:!! 15 through 16

total_coefficient:!! =

Trailing_one:!! .,

-1 -1

<DNewloMATLAB?WM(hthis� ��OfreMl�.

Elap:!Ied t� ill 0.'459269 second!!. IT;, »

B. 2) CAVLC Decoder

® NrwtoMATLA8?Wat<:hthi< 'liIIw. � � orfe6d �

612

�MATLAB1WIltChthil:ildm W<!' l2tinm. O" " ad�.

ColUDUlII 1- Chroug" 1-1

I:l. .. p .... d time ill 0.165179 .. "cond ... Ix. »

REFERENCES [I] Jun-Young Lee, Jae-Jin Lee, and SeongMo Park, "New Lookup Tables

and Searching Algorithms for Fast H.264/AVC CAVLC Decoding" , IEEE Transaction on circuits and systems for video Technology, Vo1.20,NO.7, JULy 2010.

[2] White paper on H.264 I MPEG-4 Part 10: Variable Length Coding, www.vcodex.com.

[3] N. Keshaveni S. Ramachandran and K.S. Gurumurthy " Implementation of Context Adaptive Variable Length Coder for H.264 Video Encoder" International Journal of Recent Trends in Engineering, Vol 2, No. 5, November 2009.

[4] Y.H. Moon, G.Y.Kim, and J.H. Kim, "An Efficient Decoding of CAVLC in H.264/AVC Video Coding Standard", IEEE Trans. On Consumer Electronics, Vo1.51, No3, August 2005.

[5] ITU-T Recommendation H.264 and ISO/IEC 14496-10, Advanced Video Coding for Generic Audiovisual Services, May 2003.

[6] ITU-T Recommendation H.264 approved by ITU-T Study Group 16 (2001-2004) under the ITU-T Recommendation A. 8 procedure on 30 May 2003.