JPEG2000算法中基于有界输入有界输出(BIBO)增益控制的小波变换定点实现技术

A Fixed-point Technique of Wavelet Transform Base on Bounded Input Bounded Output(BIBO) Plus Value in JPEG2000 Algorithm

该文提出一种在JPEG2000算法中兼容5/3小波变换和9/7小波变换高效硬件定点实现技术。所提出的技术使用9/7提升小波变换的有界输入有界输出(Bounded Input Bounded Output,BIBO)增益来确定小波变换中间值的存储位深,使用5/3提升小波变换的BIBO增益来确定9/7提升小波变换中量化参数的选择方式和量化的实现方式,最终使用同一存储空间来存放定点5/3提升小波变换和定点9/7提升小波变换系数。该文提出的技术不仅大大节省了JPEG2000算法中小波实现模块中的硬件存储资源和算法计算量,而且也节省了后续基于上下文的位平面算术编码模块和率失真优化截取模块的存储资源和算法计算量。

A new efficient technique using the compatible hardware framework to realize the integer 5/3 wavelet transform and 9/7 wavelet transform is proposed. The bit depth of the temp coefficients in the wavelet transform is based on the Bounded Input Bounded Output （BIBO） plus value of 9/7 wavelet transform, at the same time, the quantization value of 9/7 wavelet transform and how to do quantization is decided by BIBO puls value of 5/3 wavelet transform. Finally, both the coefficients of 5/3 wavelet transform and 9/7 wavelet transform are saved in the same memories. The proposed technique not only saves the hardware memory resource and computational complexity of the wavelet transform module, but also saves the hardware memory resource and computational complexity of the bitplane arithmetric coding based on the context module and post-compression rate-distortion optimization module.