HIGHLY SCALABLE IMAGE COMPRESSION BASED ON SPIHT FOR NETWORK APPLICATIONS
Habibollah Danyali and Alfred Mertins
School of Electrical, Computer and Telecommunications Engineering University of Wollongong, Wollongong, NSW 2522, Australia
Email: hd04, mertins @uow.edu.au
In general, an level wavelet decomposition allows at most
levels of spatial resolution. To distinguish between different resolution levels, we denote the lowest spatial resolution level as level . The full image then becomes resolution level 1. The three
subbands that need to be added to increase the resolution from level to level
are referred to as level
resolution subbands (see Fig. 1). An algorithm that provides full spatial scalability would encode the different resolution levels separately, allowing a transcoder or the decoder to directly access the data needed to reconstruct with a desired spatial resolution. The original SPIHT algorithm, however, encodes the entire wavelet tree in a bitplane by bitplane manner and produces a bitstream that contains the information about the different spatial resolutions in no particular order.
I - 217
0-7803-7622-6/02/$17.00 ©2002 IEEE
2. HIGHLY SCALABLE SPIHT (HS-SPIHT)
Traditional image coding systems have only focused on efﬁcient compression of image data. The main objective of such systems is optimizing image quality at given bit rate. Due to the explosive growth of the Internet and networking technology, nowadays a huge number of users with different capabilities of processing and network access bandwidth can transfer and access data easily. For transmission of visual data on such a heterogenous network, efﬁcient compression itself is not sufﬁcient. There is an increasing demand for scalability to optimally service each user according to his bandwidth and computing capabilities. A scalable image coder generates a bitstream which consists of a set of embedded parts that offer increasingly better signal-to-noise ratio (SNR) or/and greater spatial resolution. Different parts of this bitstream can be selected and decoded by a scalable decoder to meet certain requirements. In the case of an entirely scalable bitstream, different types of decoders with different complexity and access bandwidth can coexist.
Over the past decade wavelet-based image compression
schemes have become increasingly important and gained
widespread acceptance. Because of their inherent multiresolution signal representation, wavelet-based coding schemes have the potential to support both SNR and spatial scalability. For efﬁcient coding of wavelet coefﬁcients, Shapiro  introduced the Embedded Zerotree Wavelet (EZW) coding scheme based on the idea of grouping spatially related coefﬁcients at different scales to trees and efﬁciently predicting zero coefﬁcients across scales. An improved scheme, called Set Partitioning in Hierarchical Trees (SPIHT), was developed by Said and Pearlman . This coder uses the spatial orientation trees shown in Fig. 1 and partitions them as needed to sort wavelet coefﬁcients according to magnitude. Further improvements of SPIHT have been published in [3–8]. Although the SPIHT coder is fully SNR scalable with excellent compression properties, it does not explicitly support spatial scalability
and does not provide a bitstream that can be parsed easily according to the type of scalability desired by the decoder. An improved version of the EZW algorithm that uses better
context modeling for arithmetic coding, an improved symbol set for zerotree encoding, and proper syntax and markers for the compressed bitstream to allow extracting various qualities and resolutions was reported in . However the decoder needs some additional side information to decode the bitstream. Tham et al.  introduced a new zerotree structure called tri-zerotree and used a layered coding strategy with the concept of embedded...
Please join StudyMode to read the full document