Intra Frame Coding for Advanced Video Coding Standard to reduce Bitrate and obtain consistent PSNR Using Gaussian Pulse

Intra Frame Coding for Advanced Video Coding Standard to reduce Bitrate and obtain consistent PSNR Using Gaussian Pulse Manjanaik.N1, Dr.Manjunath.R2 Electronics and Communication Engineering, Jain University, Bangalore, India. 2 Senior 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), Mode7(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.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 UnionTelecommunications 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. 1 456

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 457

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 3 458

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. M I J K L A a e i m B b f j n C c g k o D d h l p E F G H 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. 459

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 with Gaussian value. 460

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 JM 8.6 Proposed JM 8.6 Proposed 39.92 39.95 883.72 862.48 37.12 37.12 619.30 598.61 34.30 34.29 413.40 398.21 31.60 31.58 274.70 263.09 Remarks PSNR 0.01 Bitrate = -3.42 % Table II Results of Foreman QCIF Proposed algorithm Remark s JM18. Proposed 6 (H.26 4) 37.00 37.58.82 37.63 34.07 37.63 31.89 37.72 PSNR consistent JM18.6 Propos (H.264 ed ) 4. 254.55 312.96 194.86 218.40 147.69 148. 133.34 Bitrate = -30.04 % 461

Table III Results of Suzie QCIF Hung et al [2008] algorithm JM 8.6 Proposed JM 8.6 Proposed 40.59 40.62 700.97 691.51 37.80 37.79 464.08 457.59 35.20 35.21 8.23 2.11.94.93 176.79 173.70 Remarks PSNR 0.02 Bitrate = -2.11 % Table IV Results of Suzie QCIF Proposed algorithm JM18.6 Proposed JM18.6 Proposed 39.85 37.31 306.16 0. 37. 37.35 208.48 180.84 34.94 37.38 134. 131.75.79 37.50 86. 96.80 Remarks PSNR consistent Bitrate = -6.12 % Table V Results of Foreman CIF Hung et al [2008] algorithm JM 8.6 Proposed 40.25 40.27 37.77 37.76 35.18 35.18.87.86 Remarks PSNR 0.01 JM 8.6 Proposed 2699.38 2659.59 1777.41 1746.19 1121.13 1102.18 703.62 694.30 Bitrate = -1.75 % Table VI Results of Foreman CIF Proposed algorithm JM Proposed JM18.6 Proposed 18.6 37.00 37.58 4. 254.55.82 37.63 312.96 194.86 34.07 37.63 218.40 147.69 31.89 37.72 148. 133.34 Remarks PSNR consistent Bitrate = 18.62% 462

Table VII Results of Tempete CIF Hung et al [2008] algorithm JM 8.6 Propos ed 39.46 39.49.11.12.73.74 29.63 29.63 Remar ks PSNR 0.02 JM Propose 8.6 d (H.26 4) 5547. 5502.9 19 3 4116. 4087.8 46 0 43. 11.5 07 8 1849. 1835.5 85 6 Bitrate = -1. % Table VIII Results of Tempete CIF Proposed algorithm Remar ks JM18.6 38.67 Propos JM18.6 Propose ed d 35.35 2370.3 13.0 2 5 35. 35.37 1743.7 1104.7 6 2 31.94 35.41 1217.2 877.35 8.98 35.48 805.20 694.70 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 (,, and ) 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. 463

Table IX Performance comparison proposed method with previous work Sequence QCIF Method Parameters 30 Bit rate Reduction Hung et al [2008] Inference -30.04 % -3.42 % Bit rate reduction JM 18.6 Foreman QCIF Suzie QCIF Average Foreman CIF 37.00.82 34.07 31.89 4. 312.96 218.40 218.40 Proposed 37.58 37.63 37.63 37.72 254.55 194.86 147.69 133.34 JM 18.6 39.85 37. 34.94.79 306.16 208.48 134. 86. Proposed 37.31 37.35 37.38 37.50 0. 180.84 131.75 96.80 JM 18.6, Intra frame, frame rate 30 frames per second CAVLC Bit rate reduction Proposed JM 18.6 Proposed Tempete CIF JM 18.6 Proposed 39.88 1256.08 38.98 914.77 38.67 2370. 35.35 13.05 37.46 870.64 38.98 673.72 35. 1743.76 35.37 1104.72 35.04 585.60 39.03 487.64 31.94 1217. 35.41 877.35.69 389.44 39.15 358.30.98 805.20 35.58 694.70 achieved more -6.12 % -2.11 % Compared -18.08 % -2.765% With -18.62% -1.75 % Hung et al [2008] -30.18% -1. % PSNR is consistent Average JM 18.6, Intra frame, frame rate 30 frames per second CAVLC - sign indicates saving + sign indicates increasing -.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) Bitrate vs Foreman Qcif 550 Bitrate Foreman Qcif Proposed Bitrate Foreman Qcif JM 18.6 500 450 400 Bitrate (kbps) 350 300 250 200 150 100 50 0 20 25 30 35 Quantization Parameter () 40 Fig.7. Bitrate vs Quantization parameter 464

Bitrate vs Akiyo Qcif 400 Bitrate Akiyo Qcif Proposed Method Bitrate Akiyo Qcif JM 18.6 350 B itrate (kbps) 300 250 200 150 100 50 0 20 25 30 35 Quantization Parameter () 40 Fig.8. Bitrate vs Quantization parameter Bitrate vs tempete cif 00 Bitrate tempete Bitrate tempete 2200 cif Proposed Method cif JM 18.6 Bitrate (kbps) 2000 1800 1600 1400 1200 1000 800 23 25 26 27 Quantization Parameter () 29 30 31 Fig.9. Bitrate vs Quantization parameter Rate Distortion Foreman qcif 50 RD curve Foreman qcif Proposed Method RD curve Foreman qcif JM18.6 Peak Signal to Noise Ratio (db) 45 40 35 30 25 20 0 50 100 150 200 250 300 350 Bit rate(kbps) 400 450 500 550 Fig.10. Bitrate vs Quantization parameter 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.4% for CIF sequence and 18.08% for 465

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.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. 466

[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 978-0-76953119-9/2008 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 467