photo sharing and upload picture albums photo forums search pictures popular photos photography help login
Victor Engel | all galleries >> Galleries >> Images used in online forums > Decoding 5D Mark II Video Interpolation
previous | next

Decoding 5D Mark II Video Interpolation

This image was taken with a 5D Mark II pointing at an LCD
monitor screen displaying horizontal white lines. The lines
are white, one pixel thick, and spaced 5 pixels apart (four
black lines between each pair).

I was experimenting with methods to reduce this aliasing
using devices in between the camera and the subject. One of
these devices was simply a number of strands of black string,
spaced evenly to create a diffraction pattern. I recorded
video, and as soon as I removed the string, this pattern
appeared. It looks like the familiar maze-like demosaicing
pattern frequently encountered when the signal frequency
is close to Nyquist.

What's interesting to me is the distribution of red and green
with respect to the distribution of blue and green. I'd expect
them to be similar, but they are not. I also notice something
else.

In addition to the horizontal banding that's obvious, there
is vertical banding, bands of sharp vs. soft. I'm not quite
sure what to make of it, yet.

The lens used was a 135mm f/2 lens. Subject distance is about
15 feet +/- less than a foot. The monitor is a Syncmaster XL20.


other sizes: small medium large original auto
comment | share
Victor Engel27-Dec-2008 02:16
I did some careful measuring and determined the following:

The 1200 line monitor is rendered on HD as 497 lines when moire is the worst (not this image). I consider for the sake of this discussion that moire is the worst when a single color extends from the top to the bottom of the screen. If we do a little math here, we discover that 1200 monitor lines, displaying a white line every 5 lines produces 240 white lines on the screen. This is rendered into 497 HD lines. So there are 2.07 HD lines per white line displayed on the monitor.

So about every other line in HD images a white line, and the balance image the black space between the white lines. But for color, what we really care about is sensor lines.

There are about 2.925 sensor lines per HD line. (5616/1920). Multiplying this by the 2.07 HD lines per white line, we get 6.057 sensor lines per white line. But it should be exactly 6 sensor lines per white line for the color to be consistent all the way across the screen. This discrepancy amounts to 6 pixels across the height of the sensor or about 2 pixels over the height of the monitor imaged here. That is probably pretty close to the error in my measurements.