Low-delay embedded 3-d wavelet color video coding with SPIHT

In this paper, a modification of the 3-D SPIHT,

Low-Delay Embedded 3-D Wavelet Color Video Coding with SPIHTBeong-Jo Kim and William A. PearlmanElectrical, Computer and Systems Engineering Dept. Rensselaer Polytechnic Institute, Troy, NY 12180, U.S.A. Tel: (518) 276-6982, Fax: (518) 276-6261 E-mail:beongjo@ipl.rpi.edu, pearlman@ecse.rpi.edu

ABSTRACTIn this paper, a modi cation of the 3-D SPIHT, which is the 3-D extension to image sequence of 2-D SPIHT still image coding, is presented in order to allow more exibility in choosing the number of frames to be processed at one time by introducing unbalanced tree structure. Simulation shows that 3-D SPIHT with reduced coding latency still achieves coding results comparable to MPEG-2, and exhibits more uniform PSNR uctuations. In addition, extension to color video coding is accomplished without explicit rate-allocation, and can be used to any color-plane representation.11 4

1 INTRODUCTIONVideo coding achieves high coding e ciency by exploiting temporal correlation among pixels in successive frames of an image sequence. Motion compensated predictive coding (MCP) is a widely accepted coding method, which has a hybrid structure whereby subband/transform coding in the spatial domain is applied to MC prediction error signals. In fact, MCP has been employed by the several international video coding standards such as MPEG-1, MPEG-2, and H.263. Another approach is 3-D subband/wavelet coding with/without MC. They are straightforward extensions of 2-D subband/wavelet to temporal domain, and are di erent mainly in the way generating 3D subbands and coding methods applied to the subbands. A strong inter-frame correlation manifests itself as an energy compaction in the temporal low-subbands. The advantages of 3-D subband/wavelet coding scheme are the lower computational complexity comparing to MCP mehtod, none-recursive structure that limits error propagation within a certain size of video segment, and multiresolutional property for possible video scalability. Inspired by excellent results of the SPIHT (set partitioning in hierarchical trees) for image coding scheme by Said and Pearlman, there has been extensive research on zero-tree based coding for video compression. There are two main streams for video coding using zero-tree structure. One kind of scheme uses conventional predictive coding method with/without motion compensation to rst remove inter-frame correlation. Then, it applies two-dimensional discrete wavelet transformation (2-D DWT) in each residual frame to reduce spatial redundancy and set up a 2-D hierarchical structure. The other method, which seems to be more powerful and e cient, uses 3-D discrete wavelet transformation (3-D DWT) to remove spatio-temporal redundancy, compact energy in particular low spatio-temporal frequency subbands, and set up three dimensional spatio-temporal hierarchical structure to better adapt to the three dimensional nature of video. Recently, 3-D extensions, 3-D IEZW (improved embedded zero-tree coding), from 2-D I

EZW and 3-D SPIHT, from 2-D SPIHT, which have the properties of progressive transmission, fast coding/decoding, complete adaptiveness (no training), precise rate control, and very simple implementation, reported excellent results without any motion compensation method. Those results reported were better than MPEG-2's with its complicated motion compensation in terms of PSNR (peak signal to noise ratio) as well as visual quality, especially at low bit-rate. In previous 3-D SPIHT, the 9/7 bi-orthogonal Daubeachies' lter was used for 3-D DWT with the same5,8,6,7,12 2 10 9 1 11 2 11

In this paper, a modification of the 3-D SPIHT,

number (3 times) of decompositions applied to the two spatial directions and to 16 frames in the temporal direction. This constraint to three levels of decomposition is the most that can be applied to the temporal direction, so that limiting the spatial levels to three may prevent further exploitation of spatial redundancy with larger size of video. Furthermore, even sixteen frame processing may cause unacceptable coding delay for transmission on some channels. In this paper, we employ a modi cation of the 3-D SPIHT tree structure to have the exibility to choose di erent numbers of decompositions between the temporal and spatial domains, in order to select a smaller number of frames to process at one time. In this way, we can decide what is the best trade-o between performance, coding delay, and memory requirements with the capability of using a larger number of decompositions in the spatial domain to compensate for possible loss of coding performance from reducing the number of frames processed at one time. With this smaller coding unit and structural freedom, there are several possible implementation options in terms of lter choice, and number of frames in one coding unit. In addition, we will consider embedded color video coding scheme, which allows implicit bit allocations among color planes, and generates one mixed color bit-stream so that we can stop decoding at any point of the bit-stream and reconstruct the color video sequence of best quality at the given bit-rate. The organization of this paper is as follows: Section 2 reviews basic principles of SPIHT. Unbalanced trees are addressed in section 3 to allow smaller number of group of frames (GOF ). Embedded color video coding scheme is discussed in section 5. Section 6 provides computer simulation results. Section 7 concludes the paper.

2 3-D SPIHT3-D SPIHT algorithm is based on three basic concepts: (1) code/transmit important information rst based on the bit-plane representation of pixels (2) ordered re nement bit-plane transmission, and (3) coding is performed along the prede ned path/trees called spatio-temporal orientation trees, which e ciently exploit the properties of a 3-D wavelet transformed video. 3-D SPIHT consists of two main stages as in 2-D SPIHT: sortin …… 此处隐藏：23120字，全部文档内容请下载后查看。喜欢就下载吧 ……