维纳滤波插值算法在运动预测中的应用

目前现有的视频编码标准 (H.2 6 3,MPEG- 4等 )都是采用半像素双线性插值的算法来对图像进行运动预测 ,为了提高视频编码的效率 ,该文在应用维纳滤波概念的基础上提出了一种新的半像素插值算法。该算法通过构造非等重的维纳滤波器来完成对图像的运动预测功能 ,它充分利用了相邻像素点间的相关性 ,同时还保持了图像中变化较大的边缘信息。实验表明 :该算法相对于双线形插值、Cubic插值、B-Spline和非线性插值在性能上有了明显的提高 ,相同比特率时其峰值信噪比可以提高 0 .6～ 1d B。

Motion-compensating predication with Wiener interpolation filters

Currently, most video coding standards (H.263, MPEG 4, etc.) perform motion prediction using half pel bilinear interpolation. The video coding efficiency was improved using a novel Wiener interpolation filter for motion compensated predication in a hybrid video codec. The proposed method introduces an unequal weighting polynomial technique to exploit the similarities of neighboring pixels in an image without loss of edge information. Simulations show that the proposed interpolation filter has less rate distortion than conventional filters. Furthermore, the results show that the 6 tap Wiener filter can more accurately interpolate the image with a coding gain of 0.6~1 dB compared to the H.263 codec for most real sequences.