2906 IEICE TRANS. COMMUN., VOL.E95–B, NO.9 SEPTEMBER 2012 PAPER Memory Efficient Set Partitioning in Hierarchical Tree (MESH) for Wavelet Image Compression Farid GHANI †a) , Member, Abdul KADER †b) , Ekram KHAN ††c) , and Badlishah AHMAD †d) , Nonmembers SUMMARY This paper presents a memory efficient version of set par- titioning in hierarchical tree (SPIHT). The proposed coder called Mem- ory Efficient SPIHT (MESH) uses a single re-usable list instead of three continuously growing linked lists as in conventional SPIHT. The list is re- initialized at the beginning of each bit-plane (coding pass) and is exhausted within the same bit-plane. Another feature of the proposed coder is that it uses a single pass for each bit-plane by merging the sorting and refine- ment passes of the conventional version of SPIHT. The performance of the proposed coder is measured in terms of coding efficiency, and the worst case dynamic memory requirements due to the list entries in each bit-plane. Performance comparison with SPIHT shows that the proposed coder re- duces the dynamic memory requirement by about 50–70% compared to the SPIHT while retaining its coding efficiency. key words: coding efficiency, image coding, reduced memory, SPIHT, wavelet transform 1. Introduction Real time transmission of images through handheld mo- bile/portable devices (with limited memory, processing power and battery life) require an image coding algorithm that can compress images efficiently with reduced compu- tational complexity. The contemporary image coding meth- ods support a wide range of functionalities but either with increased computational complexity (e.g. JPEG2000) [1], or increased memory requirement (e.g. SPIHT [3]). Also, browsing of images over Internet from image data sets re- quires fast encoding and decoding speed with a better rate distortion performance. Combined with the progressive pic- ture build up of the wavelet based coded images, good qual- ity images at the earlier stages of the transmission is becom- ing an important element of these type of codecs. This is particularly important, if the web pictures are browsed over the wireless lines, where limited channel capacity, storage limitations and computational complexity of portable multi- media devices will be the decisive factors. The dyadic wavelet decomposition of images gives ex- cellent energy clustering in space and frequency. The suc- cess of a wavelet-based image coding algorithm depends on Manuscript received October 30, 2011. Manuscript revised April 14, 2012. † The authors are with School of Computer and Communica- tion Engineering, Universiti Malaysia Perlis, 02000 Kuala Perlis, Perlis, Malaysia. †† The author is with the Department of Electronics Engineering, Aligarh Muslim University, Aligarh, 202002, India. a) E-mail: faridghani@unimap.edu.my b) E-mail: kdr2k4@rediffmail.com c) E-mail: ekhan67@gmail.com d) E-mail: badli@unimap.edu.my DOI: 10.1587/transcom.E95.B.2906 the exploitation of this property of the wavelet transform. Over the years, a number of very successful wavelet-based image coding algorithms have been proposed. Among them, the zerotree based coders are very widely used [2]. The ze- rotree algorithms work by grouping the wavelet coefficients corresponding to the same spatial location and orientation to form a spatial orientation tree (SOT). A zerotree is then a SOT with no significant coefficient with respect to the given threshold. The wavelet encoders proposed in [2], [3] are based on this concept. Among these, the Set Partitioning In Hierarchical Trees (SPIHT) proposed by Said and Pearlman [3] is a fast and efficient image compression technique and is the most popular coder. It has an excellent rate-distortion performance with low computational complexity while gen- erating an embedded bit stream. Like EZW [2], SPIHT generally operates on an entire image at once. The whole image is loaded and transformed, and then the algorithm requires repeated access to all co- efficient values. There is no structural in which the coef- ficient values are accessed. The random coefficient access requirement of the SPIHT algorithm hinders its use in cer- tain memory-constrained environments. The capability to encode a large image without storing the entire image in memory was an important feature in the JPEG2000 requirements specifications [1]. Though SPIHT [3] has excellent rate-distortion performance, but the use of pixel/set lists that significantly increased dynamic memory requirement, limits its uses in memory constrained multi- media applications such as image transmission in wireless sensor networks. Thus, there is a need for reducing the dy- namic memory used in zero-tree like algorithms. Embed- ded block coding with truncation (EBCOT) [4] algorithm which is the part of JPEG2000 standard, restricts the mem- ory usage through encoding of pixel blocks. However, it has higher complexity due to the use of adaptive arithmetic coding, multiple coding passes, and complex rate-distortion optimization. The No List SPIHT (NLS) is developed as a listless version of SPIHT [5]. It uses fixed size state table or mark- ers instead of variable size lists in SPIHT. That is the coding- time memory requirement of NLS is always fixed and is in- dependent of number of passes to be executed. This not only reduces the memory, but also makes algorithm faster as memory read/write operations are eliminated. Another extension of the SPIHT algorithm in memory constrained environment is proposed by Pearlman [6] in which coeffi- cients are grouped in many small spatial blocks in such a Copyright c  2012 The Institute of Electronics, Information and Communication Engineers