【摘要】 为提高噪声模型的估计精度,改善系统率失真性能,文中提出了一种基于残差子带分组聚类的自适应噪声模型估计方法.首先根据频率高低对残差子带进行分组,然后由组内子带残差样本生成特征矢量,进而利用改进的模糊c-均值聚类算法对当前解码子带进行聚类,最后计算出每类残差系数的噪声参数.实验结果表明,相比于相邻子带聚类-方差估计算法,文中所提算法能够更加准确地匹配残差分布特征,率失真性能平均提升0.60 d B,且解码时间平均节省40.59%.更多还原

【Abstract】 In order to improve the estimation accuracy of noise model and the rate-distortion performance of the system,an adaptive noise model estimation method on the basis of residual sub-band grouping is proposed. In this method,firstly, residual sub-bands are grouped according to their frequencies. Secondly, feature vectors are generated from the residual coefficients of all sub-bands in the same group. Then,the coefficients in each subband are clustered into different classes by means of improved fuzzy c-means clustering. Finally,the noise parameters of each class of residual coefficients are estimated successfully. Experimental results show that,in comparison with the method on the basis of adjacent sub-band clustering and variance estimation,the proposed method matches the residual distribution characteristics more accurately,improves the average rate-distortion performance by 0. 60 d B,and saves the decoding time by 40. 59%.更多还原