Deinterlacing
Deinterlacing the process of converting interlaced images (known as fields)
into non-interlaced form (comprised of frames).
Interlaced video draws only half of the lines on the screen for each frame
(alternately drawing the odd and even lines for each frame), taking advantage of
the time it takes for a image to fade on a CRT to give the impression of double
the actual refresh rate, helping to prevent flicker.
When displaying video on a display that can support a high enough refresh
rate that flicker isn't perceivable, interlaced video can be deinterlaced for
better viewing. When a display cannot interlace but must draw the entire screen
each time, the video must be deinterlaced before it can be displayed. All
displays except for CRT screens must deinterlace.
Deinterlacing Methods
There are various methods to deinterlace video, each producing different
artifacts. Artifacts will always be present in deinterlaced video, as the
process must deal with the fact that interlaced video has only half the
information of non-interlaced video.
There are two basic methods of deinterlacing: combination, where the even and
odd frames are combined into one image and then displayed, and extension, where
each frame (with only half the lines) is extended to the entire screen.
Field Combination Deinterlacing
- Weaving is done by adding consecutive fields together. This is fine when
the image hasn't changed between fields, but any change will result in
artifacts known as "mouse teeth", when the pixels in one frame do not line
up with the pixels in the other, forming a jagged edge. This technique
retains full vertical resolution at the expense of half the temporal
resolution.
- Blending is done by blending, or averaging consecutive fields to be
displayed as one frame. The mouse teeth are avoided because both of the
images are on top of each other. This instead leaves an artifact known as
ghosting. The image loses vertical resolution and temporal resolution. This
is often combined with a vertical resize so that the output has no numerical
loss in vertical resolution. The problem with this is that there is a
quality loss, because the image has been downsized then upsized. This loss
in detail makes the image look softer.
- Selective blending, or smart blending or motion adaptive blending, is a
combination of weaving and blending. As areas that haven't changed from
frame to frame don't need any processing, the frames are weaved and only the
areas that need it are blended. This retains full vertical resolution, half
the temporal resolution, and has fewer artifacts than weaving or blending
because of the combination of them both.
Frame Extension Deinterlacing
- Half-sizing displays each interlaced frame on its own, resulting in a
video with half the vertical resolution of the original, unscaled.
Understandably, this is never used for regular viewing.
- Line doubling takes the lines of each interlaced frames (consisting of
only even or odd lines) and doubles them, filling the entire frame. This
results in the video having effectively half the vertical resolution, scaled
to the full resolution. While this prevents mouse teeth, it causes a
noticeable reduction in picture quality. This technique is also called bob
deinterlacing, because the fields are bobbed up and down.
Both frame combination and frame extension lend themselves to a method called
motion compensation. Deinterlacers that use this technique are often superior
because they can use information from many fields, as opposed to just one or
two. For example, if two fields had a person's head moving to the left, then if
it weaving was applied, mice teeth would appear. If blending was applied,
ghosting would appear. Selective blending would also create ghosting. Both of
the frame extension methods would have no artifacts, but the level of detail on
the face would be half. Motion compensation (ideally) would see that the face in
both frames is the same, just transposed, and would combine the face (through
weaving or some other more advanced method) to get full detail in both output
frames. This needs to be combined with a scene change detection algorithm,
otherwise it will attempt to find motion between two completely different
scenes.
The best deinterlacers combine all of the methods mentioned above. The fields
are bobbed, so the framerate is then kept. Motion compensation is done. In the
areas that it cannot find a motion match, it falls back on selective blending.
Where deinterlacing is done
Deinterlacing can be done (if it needs to be) at various points in the chain
from filming to watching. When it is done affects the quality of the deinterlace,
because the quality of the deinterlacer can vary.
- If it is done in the filming studios, it should be done very well. The
people doing it are professionals, and have minimal time constraints. They
should also have access to expensive and powerful deinterlacers.
- If it is done at the time of broadcasting, the quality of the
deinterlace can vary. It should be organised by professionals, who have a
reasonable budget and powerful processors. On the other hand, it needs to be
done in real time, so the effort that the deinterlacer can put in is limited
by time.
- If it is copied onto a computer and processed there, the quality can
also vary immensely, yet (theoretically) a high quality level should be
possible, because there are no restrictions on time and there are some very
good, free, deinterlacers. However, many people who do this do not know much
about deinterlacing, and when making a choice about which deinterlacer and
settings to use, will make a random decision.
- If done by an embedded electronic device, the quality varies depending
on the overall quality of the device. High-quality electronic devices also
have high-quality deinterlacers.