Analysis of Integer Transformation and Quantization Blocks using H.264 Standard and the Conventional DCT Techniques

Transcription

1 Priyanka P James et al, International Journal o Computer Science and Mobile Computing, Vol.3 Issue.3, March- 2014, pg Available Online at International Journal o Computer Science and Mobile Computing A Monthly Journal o Computer Science and Inormation echnology IJCSMC, Vol. 3, Issue. 3, March 2014, pg RESEARCH ARICLE ISSN X Analysis o Integer ransormation and Quantization Blocks using H.264 Standard and the Conventional DC echniques Abstract:- Priyanka P James 1, Chirappanath B Albert 2, Inbanila.K 3 Christ University, Bangalore, 60 priyankajames1989@gmail.com albertbazil@gmail.com inbanila.k@christuniversity.in H.264 standard, transormation is a technique o converting the image samples into elementary requency components. Integer ransormation helps in removing redundant data rom an image and involves only real components and quantization reduces the precision o transorm coeicients. H.264 is a lossy compression ormat because o Integer ransormation and Quantization. his paper deals with the understanding and the analysis in the reduction o complexity o integer transormation and quantization blocks using H.264 and the conventional techniques. Keywords H.264 standards, Quantization, Compression, Integer ransormation I. INRODUCION he image or video CODEC converts the image or the motion-compensated residual data into another domain, the transorm domain[4]. he choice o transorm depends on a number o criteria like 1. Data in the transorm domain should be de-correlated, i.e. separated into components within minimal interdependence, and compact, i.e. most o the energy in the transormed data should be concentrated into a small o values. 2. he transorm should be reversible. 3. he transorm should be computationally tractable, that is it should be o low memory requirement, achievable using limited-precision arithmetic, low number o arithmetic operations, etc. Discrete Cosine ransorm (DC) operates on blocks o N N image or residual samples and hence the image is processed in units o a block[3]. Block transorms have low memory requirements and are well suited to 2014, IJCSMC All Rights Reserved 873

2 Priyanka P James et al, International Journal o Computer Science and Mobile Computing, Vol.3 Issue.3, March- 2014, pg compression o block-based motion compensation residuals but tend to suer rom artiacts at block edges ( blockiness ). Image-based transorms operate on an entire image or rame or a large section o the image known as a tile [2]. he Discrete Cosine ransorm (DC) operates on X, a block o N N samples, typically image samples or residual values ater prediction, to create Y, an N N block o coeicients[6]. he action o the DC can be described in terms o a transorm matrix A. he orward DC (FDC) o an N N sample block is given by: Y AXA (1.1) Where X is a matrix o samples, Y is a matrix o coeicients and A is an N N transorm matrix. he elements o matrix A are: where 1 2 C i ( i 0), Ci ( i 0) N N xy x y ij i0 j0 (2 j1) i Aij Ci cos (1.2) 2N N1N1 (2 j 1) y (2i 1) x Y C C X cos cos 2N 2N DC o previous video coding standards provide transormation but produced inverse transorm mismatch problems, due to loating point DC. In order to overcome the mismatch problems, H.264 standard deals with integer transorm[4]. II. H.264 INEGER RANSFORMAION AND QUANIZAION H.264 integer transorm is ree rom multipliers and it deals with additions and shits, which makes it low complex. In earlier standards, there was an obvious boundary between the transorm, converting a block o image samples into a dierent domain, and quantizations, reducing the precision o transorm coeicients [2]. his boundary is less obvious in an H.264 codec, with an overlap o the transorm and quantization stages. his, together with the new approach o exactly speciying a reversible integer transorm core, makes the H.264 transorm and quantization stage signiicantly dierent rom earlier compression standards [3]. he previous video coding standards relied on Discrete Cosine ransorm (DC) that provided the transormation but produced inverse transorm mismatch problems.h.264/mpeg-4 part 10, uses an integer transorm with a similar coding gain as a 4x4 DC. It is multiplier-ree, involves additions, shits in 16-bit arithmetic, thus minimizing computational complexity, especially or low-end processes [1]. Integer transorm is achieved by: Using a core transorm, an integer transorm, that can be carried out using integer or ixed point arithmetic Integrating a normalization step with the quantization process to minimize the number o multiplications required to process a block o residual data [1]. he scaling and inverse transorm processes carried out by a decoder are exactly speciied in the standard so that every H.264 implementation should produce identical results, eliminating mismatch between dierent transorm implementations [5]. (1.3) 2014, IJCSMC All Rights Reserved 874

3 Priyanka P James et al, International Journal o Computer Science and Mobile Computing, Vol.3 Issue.3, March- 2014, pg Figure.1. Development o integer transorm and quantization [1] Consider a block o pixel data that is processed by a two-dimensional Discrete Cosine ransorm (DC) ollowed by quantization, i.e. rounded division by a quantization step size, Q step. Y AXA (2.1) a a a a b c c b A a a a a c b b c (2.2) Where, 1 1 a, b cos( ) c cos( ) C (2.3) A C R (2.4) 4 Where, R (2.5) Y C 4 R 4 X C 4 R 4 C 4 X C 4 R 4 R 4 (2.6) C X C S , IJCSMC All Rights Reserved 875

4 Priyanka P James et al, International Journal o Computer Science and Mobile Computing, Vol.3 Issue.3, March- 2014, pg Where, S R R (2.7) Scale the quantization process by a constant (2 15 ) and compensate by dividing and rounding. Combine S4 and the quantization process into M4. M S 2 Q step 15 (2.8) M 4 m( QP%6, n) / 2 loor ( QP/6) (2.9) m( QP%6, n) Y round C 4X C 4 loor ( QP/6) (2.10) Multiplication by two can be perormed either through additions or through let shits, so that no actual multiplication operations are necessary. hus, the transorm is multiplier-ree [2]. III. MAHEMAICAL CALCULAIONS 1.DC and Quantization For the mathematical calculations o the DC and the quantization the matrix A is taken as deined by the given standards. Assuming the matrix X is assumed ransormed and quantized output matrix A Y AXA X AX Y A X A QP 6 Q step AX A Y1 Qstep , IJCSMC All Rights Reserved 876

5 Priyanka P James et al, International Journal o Computer Science and Mobile Computing, Vol.3 Issue.3, March- 2014, pg Integer DC and Quantization. In the case o Integer DC and Quantization technique the predeined matrix C is given below and the values o C are integres reducing the complexcity o the whole procedure. he matrix X is taken to be the same as in the previous case. In here the inal values are rounded o. m( QP%6, n) Y round C 4X C 4 loor ( QP/6) C X C X C QP m(0, n) C X. C m(0, n) Y IV. SIMULAION RESULS Figure.2. DC and quantization output 2014, IJCSMC All Rights Reserved 877

6 Priyanka P James et al, International Journal o Computer Science and Mobile Computing, Vol.3 Issue.3, March- 2014, pg V. CONCLUSION Figure.3.Integer DC and Quantization output From the above mathamatical calculations and the simulation results it is made clear that the H.264 gives us the most identitical compression perormance to the DC. he simulation o Integer DC and Quantization output shown above is the stage beore the rounding o is carried out. he interger DC and quantization technique the output matrix Y is made simpler and reduced in complexicty as compared to the normal DC and quantization technique only by the usage o basic mathematical operations like addition and substraction along with some phase shiters. REFERENCES [1] Iain E. Richardson, he H.264 Advanced Video Compression Standard, 2 nd edition,vcodex limited, UK, 2010 [2]homas Wiegand, Gary J.Sullivan, Gisle Bjonteguard and Ajay Luthra, Overview o the H.264/AVC Video Coding Standard,IEEE ransactions on Circuits and Systems or Video echnology, vol. 13, no. 7, July [3] ien- Ying Kuo and Chen-Hung Chan, Fast Variable Block Size Motion Estimation or H.264 Using Likelihood and Correlation o Motion Field, IEEE ransactions on Circuits and Systems or Video echnology, Vol. 16, no. 10, October [4] A.Ahmed,N.Khan.S.Masud,MA.Moud,Perormance Evaluation o Advanced Features o H.26L Video Coding Standards. Proceedings IEEE INMIC 2003 [5] H. Schwarz, D. Marpe and. Wiegand, Analysis o Hierarchical B Pictures and MCF, IEEE International Conerence on Multimedia and Expo (2006), pp [6] P. List, A. Joch, J. Lainema, G. Bjontegaard and M. Karczewicz, Adaptive Deblocking Filter, IEEE ransactions on Circuits and Systems or Video echnology, vol. 13, no. 7, July , IJCSMC All Rights Reserved 878