Abstract: The transform-domain Wiener filter can not denoise efficiently in the highly improved H.264/AVC video encoder. This paper improves the calculation of filter factors of DCT coefficients by modifying the normalized covariance estimator according to its high correlation between the block data. After comparison with other existed denoising algorithm under the same encoding circumstances, the improved transform-domain Wiener filter algorithm has higher compression ratio and averaged peak signal-to-noise ratio, and can denoise efficiently and stably even under increased noise intensity circumstances.

Key words: transform-domain Wiener filter, normalized covariance, AC coefficient

摘要： 针对变换域维纳滤波算法在低配置T264编码环境下降噪性能存在急剧下降的现象，提出在计算DCT系数归一化协方差矩阵时，利用块内数据间的强相关性，对交流系数的方差值进行估值修复，进而改进DCT系数调节因子的计算式。在已知噪声强度和低配置T264编码环境下，通过与现有算法做对比实验，证明对于低慢速和画面比较平滑的视频序列，改进的变换域维纳滤波在压缩效率、平均峰值信噪比两个性能评价指标上都优于其他几种算法，而且在噪声强度增强的情况下仍具有相对稳定的降噪能力。

关键词: 变换域维纳滤波, 归一化协方差矩阵, 交流系数