基于整型提升小波变换的图像处理及DSP实现

针对数字信号处理器(DSP)的并行特性,提出了一种二维整型提升小波的并行体系结构。该结构采用乒乓缓存策略,使得数据的传输和小波变换能够同时进行;用基于行的列变换方法使得列变换只需少量行变换结果就能进行列变换;用移位操作代替乘法操作,大大减少了算法的运算量。整个结构采用流水线设计,提高了硬件资源的利用率和降低了算法的中间存储量,实现了图像的实时小波变换。实验证明,该算法与原算法相比,速度提高了15倍,达到了每秒85帧;重构图像的峰值信噪比(PSNR)虽然比离散小波变换要低一些,但仍达到了42 dB以上,因此该算法具有广阔的应用前景。

Image processing based on integral lifting scheme and its implementation by DSP

Based on the parallel feature of Digital Signal Processor(DSP),the parallel architecture of 2-D integral lifting scheme was presented to perform simultaneously data transfer and wavelet transform with the strategy of ping-pong buffer.The column transform method based on line transform permits column transform in a few line transform results only;multiplication operations are substituted by shift operations to enhance the transform speed.The whole architecture is optimized in the pipelined design way to achieve higher hardware utilization and decrease the intermediate data quantity.The transform speed of optimized algorithm is 85 frame per second and 15 times faster than original algorithm's.The Peek Signal-to-noise Ratio(PSNR) of optimized algorithm is lower than original algorithm's,but it is over 42 dB,thereby optimized algorithm has high application value.