Non-Memory-Based and Real-Time Zerotree Building for Wavelet Zerotree Coding Systems

The wavelet zerotree coding systems, including Embedded Zerotree Wavelet (EZW)[1] and its variants Set Partitioning In Hierarchical Trees (SPIHT)[2] and Space Frequency Quantization (SFQ)[3][4], have three common procedures: 1) 2-D Discrete Wavelet Transform (DWT)[5], 2) zerotree building and symbol generation from the wavelet coeÆcients (illustrated in Figure 1), and 3) quantization of the magnitudes of signi cant coeÆcients and entropy coding, where the second procedure is an important one. All recently propsed architectures ([6]-[9]) for wavlelet zerotree coding use memories to build zerotrees. In this paper we contribute to building the zerotrees in a non-memory-based way with real-time performance leading to the decrease of hardware cost and the increasement of processing rate which is especially desirable in video coding. One of our main ideas is to rearrange the DWT calculations taking advantage of parallel and pipelined processing so that any parent coeÆcient and its children coeÆcients in zerotrees are guaranteed to be calculated and outputted simultaneously.