Recently I was prompted to try taking photos of the TV which currently serves as a monitor for my XT, because reasons (which may be made clearer in the future if I'm not lazy and/or distracted by other things). After much trial and error, I think my current results provide an accurate representation of what's actually visible on the screen.
Getting there wasn't exactly a walk in the park, so perhaps this post could prove useful for someone. Here are my results, followed by how I got them, along with the usual long winded observations! (Excited yet?)
But first, a quick rundown of the ingredients:
- Video source: an IBM Color/Graphics Adapter (late-type composite output stage, AKA 'new-style') sending 60p NTSC video over the RCA jack.
- TV set: A 14" CRT Brother model (BR7414H) made in China, 1998.
- Camera: Canon EOS 700D (aka "Rebel T5i"), with a Canon EF-S 18-135mm zoom lens.
And the sample photos - click for full size:
As you can see this TV is far from perfect: it has some convergence problems (red is somewhat 'off' in the top-right area of the screen), and beam focus is quite bad around the center of the picture, especially with things like bright white text on black. But fixing the CRT is a matter for another day... perhaps. This is all about taking optimal photos, so let's get on with the details:
If you only care about the actual screen content, get rid of any ambient light. You don't want white balance mismatches between your room lighting and the TV, and you certainly don't want reflections off the glass face. Any photon that isn't coming directly from the TV is TEH ENEMY, so keep 'em out.
Also, use a tripod, yes?
Obviously this changes from set to set, so use your judgment and experiment. Just don't do silly things like over-brightening the blacks, etc. You want to represent what you see on the screen during normal viewing. This would be good to keep in mind: camera settings probably give you some leeway to experiment with brightness; but with contrast and saturation you have to be a bit more careful. You might even want to decrease the saturation a bit from what looks subjectively "right" at first glance, because the clipping of colors due to oversaturation can skew the apparent color space in annoying ways.
When you're shooting a still image (as I was doing here) there's no need to match the TV's refresh rate, which is ~59.93Hz in my case. In fact 1/60 gave me terrible interference patterns (moiré), and while I have a hunch why that happens, that doesn't really matter. 1/20 worked well enough in my case.
Can be a bit tricky; you don't have to know the ¡SCIENCE! in intimate detail, just understand the focal length and the interplay between this and shutter speed/ISO. In combination with the other values listed, F = 4.0 turned out to be optimal for me.
Generally you don't want to set this very high because it can introduce noise, even though the face of a CRT TV is full of fine detail in much greater amplitude (scanlines, phosphor dots/strips) so you're unlikely to notice any of that. I settled for 200.
In theory 6500K is the standard white point for TV, and decent/newer sets can be adjusted to this (it'd even be somewhat accurate). In practice, older CRT TV color temperatures are all over the place: it was found that an electric blue tint had more of a sensory impact on consumers, so manufacturers used to race each other to the higher end of the scale (not unlike the music industry's Loudness War). So again, it's up to experimentation.
My TV appeared too blueish even with the camera's highest WB preset of 7000K, and I had to manually add a white balance shift factor of A4 (towards amber/away from blue) + G3 (towards green/away from magenta). Adobe Camera Raw interprets this as a WB of 8200K with a tint value of -8; not sure how correct those values are, but the result looks correct indeed.
Picture settings - contrast and sharpness:
The exact workings of these parameters are probably particular to Canon EOS cameras, but of course the idea is to avoid artificial 'enhancements' as much as possible. In my camera Contrast runs from -4 to 4, and only the very lowest setting gives me faithful reproduction: any higher than -4, and the CGA's dark grey gets even darker than it is on the TV screen, eventually clipping into black. I suspect (but don't know) that -4 is the 'neutral' setting, and anything higher artificially adds contrast.
Artificial sharpening is bad in general, and on a nice focused picture of a CRT screen with all those tiny details, you want it like you want a flesh eating disease. The "Standard" setting in this camera has Sharpness on 3 (it goes from 0 to 7); for best results I take it down to 2. 0 looks a little softer than it should be; I suppose it might depend on the camera's anti-alias filter and other things, but there you go.
Zoom and focus:
Unless your CRT is of the flat-screen variety, it's a bad idea to just shove the camera lens right in its face and shoot away. Up close, the convex surface introduces a very exaggerated barrel distortion to your picture (I made that mistake in a previous post). Just position the camera about 1m away and zoom in with the lens.
You shouldn't auto-focus either, because such mechanisms are going to have a terrible time 'locking' onto the confusingly-detailed picture on the TV screen. A simple but effective way to get good manual focus: on a digital LCD viewfinder, an optimally-zoomed image of a CRT TV will show prominent moiré patterns, but this only affects the image on the viewfinder itself. Find the two points of the focus ring where these patterns completely disappear from the viewfinder. At the mid-point between them, the interference will be at its worst: that's actually your optimal focus, because this means you're sampling the most high-frequency detail (therefore, the sharpest image). ¡SCIENCE! to the rescue again!
If you're shooting RAW images, the above settings aren't pemanently (destructively) applied to the raw data; they're saved as parameters to be applied later, through something like Lightroom or Adobe Camera Raw. You should also take color spaces into account. The Canon produces .CR2 files w/the Adobe RGB color space; for optimal results I do this:
- Set Photoshop's default RGB working profile to Monitor RGB (sRGB IEC61966- 2.1)
- Import the .CR2 with Adobe Camera Raw: if the camera settings were good enough, no further adjustments are necessary and they'll be applied at this point
- When prompted, convert the document's colors to the current working space
- When saving the image, include the ICC color profile just in case, so any viewer will be told up front about our nice and standard sRGB color space.
This proved trickier than anticipated. Again, those tiny little dots of phosphor light create a lot of high-frequency image data; when downscaled by an appreciable factor (e.g. for thumbnails) the frequency is increased, shoots past the Nyquist limit and you're back in moiré city.
Before you resize then, you want to remove detail with too high a frequency - in other words, apply a low-pass filter: Gaussian blur is just that. How much of it you apply (radius) is up to the image and the scale factor: find the amount that acceptably reduces interference patterns and don't go any higher than that.
But it's still not that simple. Photoshop (CS6, at least) doesn't do resizing and blurring in a gamma-aware manner: it simply blends color values linearly and calls it a day. This will considerably darken your image and shift your hues, and you don't want that. Fortunately, it seems that working in 32 bits/channel mode eliminates this problem, or at least mitigates it well enough that you can't tell it's there. So, first convert the image to 32 bits/channel, apply the blur, then convert back to 8 bits/channel (use "Exposure and Gamma" as the method). Then resize - the "Bicubic (best for smooth gradient)" filter looks nice enough for my purposes.
One issue that still affects the final photos is non-uniform brightness - the image is darker towards the edges of the screen than near the center. But I guess that's a side-effect of the convex face, so I shouldn't correct for it really... even if it was easy (it's not).
That's it then... any comments, ideas, and/or tips for further improvements are both welcome and appreciated.