基于牛顿迭代法的快速精确的反量化算法

A Fast and Precise Multimedia Inverse Quantization Algorithm Based on Newton's Method

非线性函数广泛应用于各种多媒体压缩算法中,但在定点SOC芯片上处理这些函数比较困难,严重影响多媒体编解码性能,因此须采用近似算法代替。本文考虑了定点SOC芯片处理能力、存储空间的限制和要求,提出一种基于部分查找表和牛顿迭代法相结合的算法,给出选取牛顿迭代法的高精度初始近似值的方法,仅需迭代一次就可以达到很高精度,通过误差分析,对初始近似值进行补偿,使精度进一步提高。实验结果表明,该方法需要的存储空间比完全查找表减小了85%,误差比Chih-Hsu Yen提出的线性选择初始近似值方法减小了96%。

In multimedia systems many nonlinear functions are applied,but they are implemented difficultly on fixed-point SOC. In this paper,considering the ability of fixed-point SOC ,the limitation of memory and the performance,a method based on lookup table and newton＇ s method is proposed. To decrease iteration times, a method to get the first precise approximation is also put forward. Through the error analysis ,a compensation method for the first approximation is proposed. The experiment proves that the method proposed in this paper needs less memory compared with full lookup table method and improves precision greatly compared with the method proposed by Chih-Hsu Yen.