适合DSP处理的低内存并行SPIHT算法

针对SPIHT(set partitioning in hierarchical trees)算法的编码过程具有重复运算、存储量大等问题,提出了一种适合于DSP(digital signal processors)处理的低内存并行SPIHT算法。该算法采用乒乓缓存策略,使得数据的传输和编码能够同时进行。通过引入基于行的整型提升方案,使得只需经少量行变换就能进行列变换,提高了小波的变换速度。根据DSP的并行特性和SPIHT算法的缺点,采用“改进的最大幅值求取方法”、“误差位数以及绝对零值和绝对零集合”、“最大值与零值图”和“单棵零树编码”等多种方法对其进行了改进,大大缓解了对内存的压力,减少了算法的运算量。该算法与LZC(listless zerotree coding)算法相比,重构图像的峰值信噪比相当,但速度提高了2倍,能满足一般的实时压缩要求。

Low memory parallel SPIHT algorithm for DSP

A low memory parallel SPIHT(set partitioning in hierarchical trees) algorithm for DSP(digital signal processors) is presented,because SPIHT algorithm has many repeat calculations and needs a large number of memories.The strategy of Ping-Pong buffer allows data transfer and code to perform simultaneously.Lifting scheme based line permits column transform in the case of a few line transform results,which improves wavelet transform speed.Based on parallel feature of DSP,SPIHT defects were improved by using methods "improved maximum magnitude","error bit number,absolute zero value and absolute zero set","maximum value and zero value map","single zerotree code",which reduce significantly memory requirement and complexity of SPIHT.Experiment results show that the peek signal-to-noise ratio(PSNR) of improved SPIHT is slightly lower than LZC(listless zerotree coding) and the compression speed is improved by 2 times than LZC,which can meet generic real-time compression requirement.