Frequency-domain filtering for down-conversion of discrete cosine transform encoded images

摘要

A discrete cosine transform (DCT) domain filter for lowpass filtering a high resolution code video image represented by a frequency domain coefficient value, such as a macroblock, prior to extraction of the video image in the spatial domain. The DCT filter masks or weights the DCT coefficients of the video image macroblock prior to processing by the inverse DCT. The filter may be implemented as a block mirror filter in the frequency domain, and the filter value may be combined with the IDCT coefficient value. The original motion vector of the high resolution code video image is translated because the low resolution reference image used by the decoder is not equivalent to the original high resolution image. Thus, the motion vector is scaled to reconstruct the low resolution prediction block that is upsampled to produce the original pixel and half-pixel values in the spatial domain. The upsampled prediction block is added to the DCT filtered DCT transformed pixel if the current macroblock is part of a code image that is a non-intraframe. After the motion compensation processing of the original macroblock, reconstructed macroblocks at low resolution are appropriately extracted.