An Efficient Method for DCT-domain Image Downsampling
DU Yao-gang, Cai An-ni, Sun Jing-ao · 2005
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 03dB better than Dugad’s method in terms of PNSR of reconstructed image, and better subjective quality can be observed on details of the image.