基于最优因果化提升结构及子带叠混的低内存DWT实现

Low-memory Discrete Wavelet Transform with Optimization of Causal Lifting Scheme and Subband Interleaving

基于提升结构的因果化实现及优化在将两带滤波器组转化为单进单出系统时子带系数的叠混模式,提出了一种改进的低内存需求的离散小波变换(Enhanced Low-memory Discrete Wavelet Transform,ELDWT)实现方法。相对于DWT的常规全局实现法,基于ELDWT实现的正、逆离散小波变换均具有与图像高度无关的低内存需求且不同的分解层可以使用不同的小波滤波器。相对于著名的基于行的离散小波变换实现(Line-Based Wavelet Transform,LBWT),当小波滤波器组中的滤波器长度差大于2时,ELDWT具有比LBWT更低的全局内存开销及系统时延。当采用MPEG-4标准中静态纹理压缩所采用的9/3滤波器组及典型的5层分解,相对于LBWT,基于ELDWT的2维DWT实现的内存需求降低了22.7%。

Based on the optimization of causal implementation for the lifting schemes and the interleaving mode for sub-band coefficients,this research presents an enhanced low-memory implementation of discrete wavelet transform called the ELDWT（Enhanced Low-Memory Discrete Wavelet Transform）.In comparison with the conventional global implementation of DWT,the ELDWT has the advantages that its memory budget is independent of the image height and different DWT filter banks cad be utilized in different decomposition/reconstruction levels.When the difference between the filters＇ lengths is greater than two,the ELDWT has lower memory requirement and less system latency than those of the line-based DWT（LBWT）.When a 5-level decomposition with the MPEG Default 9/3 filter bank is adopted,the overall memory is reduced by 22.7% in comparison with the LBWT.