Doom9's Forum - View Single Post

0xdeadbeef · 18th June 2003, 23:09

If CCE would get the uncompressed stream as input, it would of course produce always better or at least equal results compared to a good transcoder.
Then again, compressing an image which was already compressed by removing the higher spatial frequencies is a less ideal scenario. Removing the higher coefficients of the discrete cosinus transformation used by JPEG/MPEG creates the typical "ghost image" artifacts on sharp edges. These are caused by the missing higher coefficients which would normally extinguish the oscillation.
When recompressing this, the encoder will try to reproduce the edge including the artifacts. When transcoding, just some more coefficients are deleted. It is obvious that the latter approach not only creates less data but also a higher image quality.
This works fine for medium compressin ratios. However at some point only removing coefficients will create too many block artefacts. Then it would me wiser to recalculate the motion vetors and/or insert/delete I-frames at the right position and so on. That's were an encoder wins over a transcoder.
But for compression ratios of about 30-40% or less, I would bet on the transcoder. Please note that some of the current one click transcoders use pretty simple algorithms.

18th June 2003, 23:09	#9 \| Link
0xdeadbeef Author of BDSup2Sub Join Date: Jun 2003 Posts: 478	If CCE would get the uncompressed stream as input, it would of course produce always better or at least equal results compared to a good transcoder. Then again, compressing an image which was already compressed by removing the higher spatial frequencies is a less ideal scenario. Removing the higher coefficients of the discrete cosinus transformation used by JPEG/MPEG creates the typical "ghost image" artifacts on sharp edges. These are caused by the missing higher coefficients which would normally extinguish the oscillation. When recompressing this, the encoder will try to reproduce the edge including the artifacts. When transcoding, just some more coefficients are deleted. It is obvious that the latter approach not only creates less data but also a higher image quality. This works fine for medium compressin ratios. However at some point only removing coefficients will create too many block artefacts. Then it would me wiser to recalculate the motion vetors and/or insert/delete I-frames at the right position and so on. That's were an encoder wins over a transcoder. But for compression ratios of about 30-40% or less, I would bet on the transcoder. Please note that some of the current one click transcoders use pretty simple algorithms.