Image Processing Reference
Compression Is About Trade-Offs
Compressing video is all about making the best compromises possible without giving up
too much quality. To that end, anything that reduces the amount of video to be encoded
will help reduce the overall size of the finished output file or stream .
Compression is not only about keeping overall file size small. It also deals with opti-
mizing data throughput—the amount of data that will steadily move through your play-
back pipeline and get onto the screen. If you don't compress the video properly, it will not
fit the pipe and therefore cannot be streamed in real time.
Reducing the number of frames to be delivered helps reduce the capacity required,
but the motion becomes jerky and unrealistic. Keeping the frame count up may mean you
have to compromise on the amount of data per frame. That leads to loss of quality and a
blocky appearance. Judging the right setting is difficult, because certain content com-
presses more easily, while other material creates a spike in the bit rate required. That spike
can be allowed to momentarily absorb a higher bit rate, in which case the quality will stay
the same. Alternatively, you can cap the bit rate that is available. If you cap the bit rate, the
quality will momentarily decline and then recover after the spike has passed. A good
example of this is a dissolve between two scenes when compressed using MPEG-2 for
broadcast TV services operating within a fixed and capped bit rate.
First We Have to Digitize
Although some compression can take place while video is still in an analog form, we only get
the large compression ratios by first converting the data to a digital representation and then
reducing the redundancy. Converting from analog to digital form is popularly called digitiz-
ing. We now have techniques for digitally representing virtually every thing that we might con-
sume. The whole world is being digitized, but we aren't yet living in the world of The Matrix .
Digitizing processes are normally only concerned with creating a representation of a
view. Video structure allows us to isolate a view at a particular time, but unless we apply
a lot more processing, we cannot easily isolate objects within a scene or reconstruct the 3D
spatial model of a scene.
Software exists that can do that kind of analysis, but it is very difficult. It does lead
to very efficient compression, though. So standards like MPEG-4 allow for 3D models of
real-world objects to be used. That content would have the necessary structure to exploit
this kind of compression because it was preserved during the creation process.
Movie special effects use 3D-model and 2D-view digitizing to combine artificially
created scene components and characters with real-world pictures. Even so, many meas-
urements must still be taken when the plates (footage) are shot.
Spatial compression squashes a single image. The encoder only considers that data, which
is self-contained within a single picture and bears no relationship to other frames in a