Low Memory Wavelet Transform And Its Coding

Image compression system based on discrete wavelet transform (DWT) has become an intensive research area in recent years. Due to its superior performance of energy compacting and redundancy removal, DWT has been incorporated into dozens of compression systems. Integrated with some elaborately designed codec, the compressed bit stream can have some very desirable properties, such as comparatively high PSNR, resolution scalability, fidelity progressivity, exact bit rate control and region of interest, etc. The inclusion of DWT in the newest image compression standard of JPEG2000[5-7] has triggered an ongoing effort to improve the implementation efficiency of DWT.However, almost all algorithms available now assume that the whole image is buffered before WT or coding works, which will be impractical if implanted to low-capacity computing platforms. The great consumption of memory by DWT or DWT-based coding has become a key bottleneck for the popularity of wavelet-based compression system. To tackle this problem, some experts have: presented low-memory and low-complexity algorithms, among which LBWT[2] is outstanding for its low memory demand and efficient implementation, and it has been included in JPEG2000.Though LBWT is very efficient in memory utilization, the coding scheme coupled with it, i.e. line-based coding (LBC) [1,2] has to use somewhat complex context information, and above all, utilize a high-complexity entropy coder-arithmetic coder[s], which is not suitable for hardware implementation on low-capacity computing platforms. In this paper, we will present a low-memory, low-complexity, hardware-friendly, yet efficient compression system based on a novel stripe-based wavelet transform proposed and an improved version of LLEC[9]This paper proposes a novel local wavelet transform technique called stripe-based VVT (SBWT). which is very suitable for parent/children tree (PCT)-based coding. Compared to the famous line-based wavelet transform (LBWT). SBWT provides less latency to PCT structure production and less total memory consumption for the whole system. Furthermore, a entropy coding method, called M-LLEC is proposed, which has low computational and memory complexity and high compression efficiency.Coupled with M-LLEC, the proposed compression scheme provides competitive high PSNR with respect to the benchmark algorithm SPIHT, while computational complexity and memory demand are greatly reduced.