Video Capture
THE
DIGITAL CORNER
Film
and print scanners with good quality are now available at quite modest prices.
But a commonly used image input device is probably the video camera and
other video sources. This is mostly due to the fact that there are so many video
cameras around. So what's wrong (or
right) with video as an image source?
Many
years ago I worked for a company that made video capture boards for personal
computers. A video capture board is
a device that converts ordinary video signals into a digital representation of
the red/green/blue values. At that
time scanners were much more expensive than they are today and we made the
argument that, while video is lower resolution than you would typically get with
a scanner, its greater flexibility more than made up for that weakness.
The obvious example was trying to get a digital image of an automobile:
If you happen to have a suitable photograph handy a scanner would be
great, but you can't really park the car on top of the scanner and hope to get
the picture you want. A less
obvious case is one where you have a photograph, but the print (or transparency)
is so small that the typical 300 DPI scanner just doesn't produce much of an
image. In fact, a scanner can be
very awkward to use just in terms of getting the right rotation on an image.
Because a video camera has adjustable optics (often including macro
capability) and provides instantaneous feedback, it is very easy to adjust the
cropping and rotation of the image before you make even one capture.
One can always process an image after scanning to change the cropping and
rotation, but such edits necessarily reduce the image quality and are generally
more time consuming than the pre-capture adjustments done with a video camera.
These
arguments are still valid today, but are weakened by the increased availability
of scanners, and also of digital still cameras.
Most "low end" digital cameras produce roughly the same
resolution as a video camera, about 640 x 480 [wow, talk
about out of date!].
A big advantage with digital cameras over video is that less processing
is needed to produce an image and so the quality is higher.
Specifically, video cameras (except for some specialty devices) produce
analog signals in which luminance is directly expressed through a voltage level,
but color information is encoded into a secondary signal with much lower
bandwidth (and thus, reduced resolution and color fidelity).
Because these luminance and chrominance components are combined into one
signal the quality is further reduced by the cross effects between the two.
That is, they can never be separated perfectly.
The advent of SVHS recording format and the "s-video" standard
improved this situation dramatically by simply keeping the two signals separate.
The luminance and chrominance signals are generated in exactly the same
way, but are never combined. Whenever
possible, use a camera with s-video output for image capture.
A
common source of confusion with video cameras is the question of resolution.
Digital still cameras are specified with resolution in pixels.
Whether or not these numbers can be believed is another issue, but at
least they are easily understood. Video
cameras are usually said to have so many "lines" of resolution. Since we all know that video images are made up of
"scan lines" it's natural to assume that this resolution figure
indicates the number of scan lines. But
this is not at all true. The number
of scan lines is determined by the video standard used:
In the U.S. the NTSC standard provides about 482 active scan lines
(another 43 lines are blanked during the vertical retrace period for a total of
525 lines). The PAL standard, used
in Europe, provides about 576 active scan lines.
While these numbers indicate an upper limit on the vertical resolution,
the number of "lines"
quoted for resolution is actually referring to horizontal
resolution. Vertical resolution is
seldom specified because we know the limits already and horizontal resolution is
the one that's harder to achieve.
So
what is this resolution figure? The
word "lines" refers to a set of black and white lines of equal spacing
printed on a card which is used to test a camera. A camera with 400 lines of resolution is one which can just
barely distinguish between the black and white lines when 400 of them are imaged
over the area of the camera's sensor. But
to make things more complicated (can't allow things to be easy now, can we?)
horizontal resolution is measured not across the full width of the sensor, but
only over a length equal to the height of the image, which is three quarters of
the width. So our NTSC camera with
400 lines of resolution is (roughly) equivalent to a pixel resolution of
533 (400 x 4/3) by 480. Since
this format is not in the correct 4:3 aspect ratio a capture system would
"over sample" to produce a 640 by 480 image, but the image quality is
still limited to no more than 533 pixels across.
Most
video capture systems use hardware circuits very similar to those found in a TV
set to decode the video signal into RGB signals.
Another approach has been quite popular in the last few years however:
A product called "Snappy" (which has been copied by several
others) is a small unit that plugs into a printer port and connects to a video
source on the other end. Rather
than use the (relatively) expensive video decoding hardware, it captures raw
video signals in digital form and then uses software to decode it to RGB values.
The tradeoff between cost and speed of processing is fairly obvious.
What's not so obvious is that the flexibility of software decoding allows
for much more sophisticated processing than can be done in hardware at a
reasonable price. Nearly perfect
filters can be implemented to separate luminance and chrominance signals.
And these filters can automatically adapt to the image content to provide
optimal quality. Other tricks can
reduce noise, maximize resolution, and improve color fidelity.
This is not to say that you should believe claims of 1280 x 960
resolution, but this technology may well be the best way to capture stills from
video with minimal expense and bother.
On
the other hand, if you can fuss with specialized equipment and want the absolute
best in video capture the thing to do is to get a monochrome camera and a set of
color separation filters (red, green, and blue filters). Three separate images can be captured, each with a different
filter, and then combined into one color image. Of course, this technique can only be used on non-moving
subjects, but motion is going to be a problem in capturing stills from video
anyway because the effective "shutter speed" is quite low at 1/30th of
a second. Solid state cameras (with
CCD sensors, for example) usually provide higher speed "shutters"
(there is no physical shutter in a video camera, but the same effect is achieved
electronically), but only for a single field at a time.
Even one frame will show blur between its two fields, and you usually
want to capture at least two full frames and average them together to reduce
noise. A monochrome camera provides
full resolution in each color rather than trading resolution for color, which
even (most) digital still cameras do.
A minor improvement in the above scheme would be to use a special video capture board which allows pixel samples to be synchronized to the camera's sensors so that there is no aliasing between the effective "sampling" of the sensor array and the sampling that occurs in converting from analog to digital. Unfortunately, the only capture device I know of which provides this feature is one I designed many years ago, and it is no longer in production.
Copyright (C) 2004 Greg Marshall
No portion of this web site may be copied or reproduced in any manner without express written consent.