一种高效的DCT域图像下采样方法

An Efficient Method for DCT-domain Image Downsampling

在许多多媒体应用中,人们往往希望压缩的视频数据可以在DCT域上进行降低空间分辨率的处理。常见的Dugad方法是从每一个8×8的DCT块中截取4×4的低频分量,但该方法容易产生错误漂移及块效应,其主要原因是没有很好地保持图像的高频分量。为此提出了一种称之为准卷积下采样方法的DCT域图像下采样方法:从8×8块DCT系数推算出16×16宏块的DCT系数,然后截取8×8的低频分量。并通过分析DCT系数矩阵的稀疏特性给出了快速算法。该下采样方法适用于任何基于DCT压缩的视频转码方案。实验结果表明,该方法重构图像的峰值信噪比比Dugad方法平均高0.3dB,并且在图像细节处,视觉效果较好。

Spatial resolution reduction of compressed video in the DCT domain are highly desirable in many multimedia applications.A parpular approach presented by Dugad is to extract the 4×4 low frequency components from each 8×8 DCT block, but this method may cause error drift and artifacts due to over-loss of high frequency components. In this paper,we propose a new method of DCT-domain image downsampling ,in which 16×16 DCT coefficients of a macroblock are deduced from the coefficients of the four blocks that constitute the macroblock,and the 8×8 low frequency components are then preserved.We also present a fast algorithm to implement this method with exploitation of the spareness of DCT matrices.The proposed method is applicable to any DCT-based compressed video transcoding. Experimental results confirm that our scheme performs about 03dB better than Dugad’s method in terms of PNSR of reconstructed image, and better subjective quality can be observed on details of the image.