Research On Efficient Pipeline Algorithms For 2DDWT With Low Memory

JPEG2000 is the new standard for image compression. The cores of JPEG2000 are DWT (Discrete Wavelet Transform) and EBCOT (Embedded Block Coding with Optimized Truncation). DWT and BPC are very computationally, as well as memory expensive operations, so improving the two cores is necessary in many application systems. After research of the JPEG2000 standard and the theory of wavelet transform, by improving the (9,7) wavelet transform and considering the relation of the two core algorithms, an efficient algorithms is proposed in this paper. The main productions can be briefly summarized as follows:1. By researching lifting-based DWT, new thought of improving work is proposed. The result of improving DWT is same the result of not-improving DWT. And improving DWT is more suitable for hardware implementation. The results of multiply and addition don't flow in fixed-bits. The two processes of predict and update is operation at the same time in one lifting step. This method reduces the length of critical path.2. By combining the thoughts of pipeline program design, new pipeline DWT architecture is proposed. The algorithms give a new design method by consider the bandwidth of memory, hardware resources efficiency, processing rate and the hardware resource quantity four aspects.3. An efficient dual-buffers architecture lifting-based two-dimensional discrete wavelet transform is proposed and implemented with high pipeline and low memory. Considering the relation between DWT and BPC, a new way of reading data from memory and the designing method of buffer are proposed. And the number of repeating reading data from memory is regarded as a performance of designing system.The researched content of this paper is based on JPEG2000 standard for image compression. The pipeline solving scheme is advanced in performance.