SNR Scalability Based on Wavelet

全文

(1)SNR Scalability Based on Wavelet Yen-Yu Chen Department of Information Managemnet. Ying-Wen Chang Department of Information Managemnet. Chungchou Institute of Technology. Chungchou Institute of Technology. miscyy@tcts1.seed.net.tw. ywchang@dragon.ccut.edu.tw. Abstract Scalable video coding methods and flexible streaming approaches are required for adapting changing network conditions in real time. This work presents scalable video coding based on a wavelet scheme. With respect to encoding, a scalable video encoder encodes predictive residual coefficients into several layers of a bit-stream. The bit-stream of the. The encoder is no longer aware of the channel capacity and does not know at which bit rate the video quality is optimal. Also, the decoder may not be able to decode all of the bit-stream received from the channel sufficiently quality enough to reconstruct the video signal. The bitstream is partially decoded at any bit rate within a range, to reconstruct a video signal with the optimal quality at that bit rate.. enhancement layer is truncated at various bit-rates,. Several streaming solutions have used variations. according to the bandwidth of the network. The. of scalable (layered) video coding methods [7-10]. residual image is reordered to form wavelet blocks.. and typically have been complemented by packet. Wavelet blocks are classified into three types, and the. loss recovery and/or error resilience mechanisms to. most significant coefficients in each type of block are. the. preserved. After the coefficients are preserved,. encountered on the Internet. In MPEG-2 [11-16] and. significant coefficients are encoded using the concept. MPEG-4 [17-18], many layered techniques [19-21],. of the significant link. With respect to decoding, the. namely SNR scalability, temporal scalability, and. bit-stream of the enhancement layer is combined. spatial scalability, are included. These technologies. with the bit-stream of the base layer to reconstruct a. code a video sequence in a base layer and an. better quality video. Simulation shows that the. enhancement layer. The enhancement layer bitstream. proposed method outperforms MPEG-4 FGS by. must be both completely received and decoded to. around 0.5 dB in terms of PSNR.. ensure that it can enhance the quality of the video.. relatively. Key words: scalable video coding, MPEG4, FGS. high. packet-loss. rate. normally. 2. Proposed Algorithm. 1. Introduction Figure 1 illustrates the main procedures. First, The real-time transmission of live video or. the positions of significant coefficients must be. stored video dominates real-time multimedia. This. determined.. Secondly,. the. most. important. investigation is concerned only with stored video. In. coefficients are preserved and the less important. an Internet streaming video system [1-6], the video. coefficients are eliminated. Finally, the coefficients. server operates between the encoder and the channel.. are encoded as a bit-stream by the zero-tree encoding.

(2) method. The bit-stream is delivered to the decoder over best-effort network. The following subsections describe each process. Wavelet Reorder residue Wavelet Block. First frame of A GOP?. N. Quantizating. Seperating layers. Encoding with significant link. Bitstream to Network. (a) Type 1. (b) Type 2. (c) Type 3. Y. Figure 2 Coefficient preservation by type. To define class block's coefficients. Generating block type map. 2.2 Quantization. Figure 1 Overview of the algorithm. The low band coefficients contribute greatly. 2.1 Reorganized and Classification wavelet block. to the PSNR of the video sequence, and are preserved Quadtree decomposition is a simple technique. using a lower quantization factor, Q1. Simulations. for representing images at various resolutions. The. results indicate that more of the larger absolute. decomposition. of. coefficients are clustered in HL and LH bands than in. coefficients of each wavelet tree into a wavelet block.. the HH band. The HH band coefficients are. The lower band coefficients are centralized in the. quantized by a larger factor, Q3, to preserve the edge. upper left of each wavelet block. information. A smaller factor, Q2, is applied in other. clarifies. the. reorganization. For three-level DWT images, all individual 8×8 wavelet blocks are stored for further processing. First, the square sum of each wavelet block in the first frame of a GOP is computed. Then, the wavelet blocks are classified into three types by the square sum: type 1 blocks have larger square sums than type 2, and type 2 have larger square sums than type 3. The square sum T is given by. places in the wavelet block to reduce the amount of transmitted information at the proper bit-rate. Another reason is that if a wavelet coefficient at a coarse scale is significant with respect to a given threshold T, then all wavelet coefficients with the same orientation at the same spatial location, on finer wavelet scales, are also likely to be significant with respect to T in the natural image but not in the wavelet residue image. 2.3 Separating Layers. N. N. T = ∑∑ block (i, j ) ……(1) 2. j =1 i =1. The advantages of layer coding are such that the residue. coefficient. is. converted. into. several. sub-streams. The coefficients are divided by a power Figure 2 depicts the method of preservation. Of type 1, all coefficients are preserved, while of types 2 and 3, one half and one-tenth of coefficients are preserved, respectively.. of 2. The data each layer are 0, 1 or –1, expressed in binary using two bits. The method efficiently reduces the data transmission overhead. A simple example is presented below..

(3) Wavelet. significant coefficient and has not been 10 0 -6 0 3 2 0 8 -5 1 -5 3 -9. coefficients L1 (23). 1 0 0 0 0 0 0 1 0 0 0 0 -1. L2 (22). 0 0 -1 0 0 0 0 0 -1 0 -1 0 0. L3 (21). 1 0 -1 0 1 1 0 0 0 0 0 1 0. L4 (20). 0 0 0 0 1 0 0 0 -1 1 -1 1 -1. After the data of each layer are separated, the. encoded, then output LINK and put children of (x,y) to LCC; Step 2.4：call ENC_LCC( ) . Step 3： encode the output symbol as bit-streams. END_enc( ). ENC_LCC( ). encoding method is applied to generate the while LCC is not empty, remove (x,y) from LCC.. transmission bit-streams.. Step 1： for △x=0,1, △y=0,1, 2.4 Significant Link. Step1.1：. The concept of significance-link is adopted and four symbols are redefined to encode the shape of a cluster: LINK, POS, NEG and ZERO. POS/NEG. if c(x+△x, ,y +△y) is a significant. coefficient and has not been encoded, then output POS/NEG and put (x,y) to LSC, go to step 1.2Otherwise output ZERO.. significant. Step1.2：if any one child of (x+△x, ,y +△y) is a. insignificant. significant coefficient and has not been. coefficient. LINK indicates the presence of a. encoded then output LINK put children of. significance-link. Two lists of coefficients are. (x+△x, ,y +△y) to LCC;. represents. the. coefficient.. ZERO. positive/negative represents. an. maintained by the algorithm: LSC (list of significant. Step 2：call ENC_LCC( ). coefficients) and LCC (list of child clusters). cn[x,y]. END ENC_LCC( ). represents the coefficient at position [x,y] in frame n. The steps of a t-level DWT residue image encoding algorithm are described as follows: Main_Enc( ) Step1：for the subband LLt, each position (x,y) of LLt . if. c[x,y] is a significant coefficient and has not. As in most image compression algorithms, the final step includes entropy coding, which employs adaptive arithmetic coding. The decoding algorithm is straightforward and can obtained by simply reversing the encoding process.. 3. Simulation Results. been encoded, then output POS/NEG and put (x,y) to LSC ; Otherwise output ZERO. Step2： for the subband LHt, for each position (x,y) belong to LHt, //LHt,HLt,HHt Step2.1： if c[x,y] is a significant coefficient and has not been encoded, then encode (x,y) output POS/NEG and put (x,y) to LSC Step2.2： output encode (x,y) ZERO Step2.3 ： if any one child of (x,y) is a. Some test sequences are used in an experiment to establish the performance of the proposed. Two 352×288 sequences, “Foreman”, and “Coastguard” are used, and a total 300 frames in YUV color format 4：2：0 are employed. In the coding scheme, the video sequences are encoded in two parts the base layer and the enhancement layer. An MPEG-4 encoder encodes these sequences in the base layer, such that the base layer bit-streams can be transmitted over a.

(4) band-limited channel. The proposed method is used. motion, motion vector is normally large and motion. to code the enhancement layer and improve visual. compensation may not suffice to stand for these. performance. The resulting enhancement layer. regions after DCT and quantization. In the base layer,. bit-streams can be transmitted at any bit-rate.. the blocking effect is very serious in these regions at. The reconstructed frames from the base layer. a low bit rate. After additional information, an. bit-streams encoded by MPEG-4 at 128K bit/s with. enhancement layer, is added, these blocking effects. TM5 rate control and only the first frames, are. are efficiently removed and PSNRs are thus. I-frames while the others are P-frames. The average. improved.. PSNR are 29.8597 dB and 26.7939 dB for Foreman and Coastguard, respectively. Notably the base layer bit-streams yield the minimum quality of bandwidth adaptation. In the sequence “Foreman”, the camera is first set on a talking man and then moved to a building. These moving frames suffer from serious blocking artifacts. In the sequence “Coastguard”, the camera moves with the boat such that the boat is always in the center of the frames. However, regions of ripple around the boat and bushes around the coast. (a). are blurred because of compression. Table 1. Simulation results of proposed method and FGS. Enh.. Foreman. Coastguard. Layer Bit-Rate. Proposed FGS Proposed FGS. 64K. 31.92 31.17 27.89 27.75. 128K. 32.19 31.89 28.98 28.52. 192K. 32.69 32.51 29.56 29.27. 256K. 33.21 33.14 29.80 29.79. For comparison, Table 1 lists the simulated results of the proposed method and MPEG4 FGS. By observation,. the. proposed. method. always. outperforms FGS. Figure 3 list the reconstructed PNSR of Foreman by adding enhancement layer at (a) 64 k bit/s (b) 128k bit/s (c) 192 k bit/s (d) 256 k bit/s. Figure 4 list the reconstructed PNSR of Coastguard by adding enhancement layer at (a) 64 k bit/s (b) 128k bit/s (c) 192 k bit/s (d) 256 k bit/s. In regions of. (b).

(5) (c). (b). (d). (c). Figure 3 Reconstructed image in Foreman (frame 255) obtained by adding enhancement layer at (a) 64 k bit/s (b) 128k bit/s (c) 192 k bit/s (d) 256 k bit/s. (d) Figure 4 Reconstructed image in Coastguard (frame 69) obtained by adding enhancement layer at (a) 64 k bit/s (b) 128k bit/s (c) 192 k bit/s (d) 256 k bit/s. (a).

(6) 4. Conclusion. [4] S.-J. Choi, J. W. Woods, “Motion-Compensated 3-D Subband Coding of Video ” IEEE Trans. on Image Process, pp. 155-167, February 1999.. This work presents a scalability algorithm based on DWT. It use significant link to preserve the relationship of significant coefficients. The more. [5] S. McCanne, V. Jackobson, and M. Vetterli, “Receiver-driven Layered Multi-cast,” Proc.. important information can be put in the front of the. SIGCOMM’96, Standford, CA, pp. 117-130, Aug.. bistream. The bistream would be truncated to adapt. 1996.. to the network bandwidth varying. Vital coefficients are always received by decoder end and the better. [6] S. McCanne, V. Jackobson, and M. Vetterli,. quality of video sequence can be reconstructed and. “Low-Complexity. play. In a simulation, the PSNR of the reconstructed. Receiver-Driven Layered Multicast,” IEEE JSAC,. video with the enhancement layers is enhanced to 2.5. vol.16, no.6, Aug. 1997,pp. 983-1001.. dB over that of the reconstructed video with only the base layer bit-stream. Moreover, the proposed. Video. Coding. for. [7] W.Tan and A.Zakhor, ”Real-Time Internet Video Using Error Resilient Scalable Compression and. method outperforms MPEG-4 FGS by around 0.5 dB. TCP-Friendly Transprot Procotol,” IEEE Trans.. in terms of PSNR.. on Multimedia, vol.1, no.2, pp.172-186, June. 5. Acknowledgement. 1999.. This research is supported partially by the National. [8] A. Eleftheriadis and D. Anastassiou, “Meeting Arbitrary QoS Constraints Using Dynamic Rate. Science Council of Taiwan under the contract. Shaping of Coded Digital Video,” in Proc. 5th Int.. number of NSC 92-2416-H-235-001. Workshop. References. Support. Network for. and. Digital. Operating Audio. and. System Video. (NOSSDAV’95), pp.95-106, Apr. 1995. [1] Dapeng Wu, Tiwei Thomas Hou, Wenwu Zhu, Ya-Qin Zhang, and Jon M. Peha, “Streaming. [9] X. Wang, and H. Schulzrinne, “Comparison of Adaptive Internet Multimedia Applications,”. Viedo Over The Internet: Approaches and. IEICE Trans. Communication., vol.E82-B, no. 6,. Directions,” IEEE Trans. on Circuit and Systems. pp. 806-818, June 1999.. for Video Technology, vol. 11, no.3, pp. 282-300. [10] Q. Zhang, G. Wang, W. Zhu, and Y. Q. Zhang,. March 2001.. “Robust Scalable Video Streaming Over Internet. [2] H.Radha,Y.Chen,“Fine-Granular-Scalable Video for. Packet. Networks,”. Packet. with Network-adaptive Congestion Control and. Video’99,. Unequal Loss Protection,” in Proc. Packet Video. Columbia University, NY, April 1999. [3] H. Radha, Y. Chen, K. Parthasarathy, R. Cohen, “Scalable Internet Video Using MPEG-4,” Signal Processing: Image Communication, pp. 95-126, September 1999.. Workshop, Kyongju, Korea, Apr. 2001. [11]. T.Sikora,. “MPEG. Digital. Video-Coding. Standards,” IEEE Signal Processing Magazine, September 1997..

(7) [12] V.Bhaskaran and K.Konstantinides, “Image and. Based. Progressive. Fine. Video Compression Standards: Algorithm and. Granularity Scalable Coding,” IEEE Image. Architectures,” Kluwer Academic Publishers,. Processing Proc. International Conference, vol. 3. 1995.. pp.556-559, 2000.. [13] ISO/IEC 13818-2 “Generic Coding of Moving Pictures and Associated Audio Information: Video,” May 1994. [14] J. Black, W. Pennebaker, C. Fogg, and D. LeGall “MPEG Video Compression Standard, Digital Multimedia Standards Series,” Chapman and Hall, 1996. [15] K. Rao and J. Hwang, “Techniques and Standards for Image, Video and Audio Coding ,” Prentice Hall, 1996. [16] B. Haskell, A. Puri, and A. Netravali “Digital Video: An Introduction to MPEG-2,” Chapman and Hall, 1997. [17] ISO/IEC/JTC1/SC29/WG11 N1642, “MPEG-4 Video Verification Model Version 7.0,” April 1997. [18] ISO/IEC JTC1/SC29/WG11 N3908 “January 2001/Pisa MPEG-4 Video Verification Model version 18.0” [19] Weiping Li, “Overview of Fine Granularity Scalablility in MPEG-4 Video Standard,” IEEE Trans. on Circuits and Systems for Video Technology, vol. 11 no. 3 pp. 301-310 March 2001. [20] Hayder M. Radha, Mihaela van der Schaar, and Yingwei. Chen,”The. MPEG-4. Fine-Grained. Scalable Video Coding Method for Multimedia Streaming Over IP,” IEEE Trans. on Multimedia vol. 3 no.1 pp. 53-68, March 2001. [21]. “DCT-Prediction. Feng. Wu,. Shipeng. Li,. Ya-Qin. Zhang,.

(8)