Image Processing Reference
Data Loss During Digital Transcoding
Compression is simply a data-reduction process. No compression is genuinely lossless
except perhaps for some of the production techniques used at the high end. In order to
compress video to usable sizes for DVD or domestic-video archiving, some consider-
able data loss has to take place. This data simply cannot be put back once it has been
Figure 3-1 shows the output of a video analyzer that is comparing the image before
and after coding and then scaling the residual error so it is visible.
More recent codecs provide additional compression tools to alleviate these effects,
which will help a great deal. The H.264 codec, for example, offers some very sophisticated
compression features, but the penalty is increased CPU power and time required to com-
press the content.
You have to decide for yourself what the acceptable level of data loss is going to be.
For film, there is no point in trying to preserve detail that is finer than the film grain,
because all you are doing is preserving noise in the original photographed image.
Mathematicians are bound to argue about this, and it is similar to the situation regarding
audio sample rates. But that is a topic for later on. For now, note the name Harry Nyquist ,
who conceived the idea of digital sampling 60 years before the technology was available
to realize his invention.
Trading off horizontal resolution for video is preferable to losing vertical resolution
due to the artifacts that vertical interpolations will introduce. If there is significant fine
detail such as text, that will become illegible after severe compression.
Figure 3-1 Residual differences after encoding (data loss).