View Single Post
Dorian Gray
Veteran Member
 
Join Date: May 2005
Location: Paris, France
 
2006-11-10, 15:20

Moogs, resolution is the accepted term for what Powerdoc was talking about. The fact that the same term is used as shorthand for pixel count in the computer world is merely confusing, especially with Bayer sensors. In fact, you've fallen victim to the "marketing hype" you so detest!

Quote:
Originally Posted by Moogs
But I still question whether the images will have the "enlargeability" (new word!) of a true 8MP DSLR shot (from a good lens) when shooting real world subjects.
This statement reveals you are fundamentally not understanding how these sensors work. An 8 MP Bayer sensor with eight million photosites cannot deliver anywhere near 8 MP of spatial resolution because the photosites are divided into red, green and blue sensors, each of which can only measure the respective colour component of the light falling on that site. A strong anti-aliasing filter, needed to eliminate the artefacts created by the interpolation process, further blurs the result. On the other hand, a 4.6 MP Foveon sensor actually delivers a genuine 4.6 MP of resolution, subject to the lens being good enough (i.e. far better than the tiny lenses on pocket digicams which are not capable of 4.6 MP of resolution).

Let me try to explain this with a simple diagram. Imagine a line of "pixels" on a Bayer sensor as follows:

R G B R G B R G B
1 2 3 4 5 6 7 8 9
(reference numbers used below)

The G "pixels" record only the green component of light falling on them (R = red, B = blue). Now let's imagine we're trying to record a purely green subject (e.g. a tree) against a perfectly white background (the sky). If the vertical green/white edge falls on the border of "pixels" number 5 and 6 (i.e. green light falls on "pixels" 1-5 and white light falls on "pixels" 6-9), one might expect the Bayer sensor to then record 1-5 as green and 6-9 as white. In reality however, the Bayer sensor has no way of knowing that 5 is green: the camera can only estimate that it might be green because 2 is reporting green and 1, 3 and 4 are all reporting a value of zero, while 7, 8 and 9 are reporting equal values and therefore might be white. Therefore the Bayer sensor can make no distinction between the vertical green/white line falling between 5 and 6, or between 6 and 7, or between 7 and 8. It can only make a distinction between it falling between 5 and 6, and falling between 8 and 9. So even though you would say my diagram has 9 pixels of resolution, the actual resolution for green against white objects is only one-third of that: i.e. 3 pixels.

A real Bayer sensor interpolates both vertically and horizontally, with a more complex RGB pattern, and so does a better job than my hypothetical example above. But the real resolution (in pixel count terms) is still less than half what the pixel count would suggest. And the anti-aliasing filter further reduces this resolution.

By contrast, here's how a Foveon sensor would look like. Each photosite ("pixel") can record the complete colour of the light falling on it:

R G B R G B R G B
B R G B R G B R G
G B R G B R G B R
1 2 3 4 5 6 7 8 9


Therefore regardless of where the green/white edge of our tree falls, the Foveon sensor will be able to record with 100% accuracy the exact edge of the tree (to the limit of the pixel count, of course). The true resolution is therefore equal to the pixel count, i.e. 9 pixels.

Yet another shorthand way to explain this is to consider the 8 MP Bayer sensor image to be a ~4 MP Foveon image that has already been enlarged (i.e. interpolated) in Photoshop by a spatial factor of the square root of 2 (i.e. 2x the pixel count).

The only case where a Bayer sensor actually delivers something close to the resolution its pixel count suggests is with a completely monochromatic subject, e.g. a test pattern of black lines on a white background. In all other cases - the real world, in other words - the resolution of the Bayer sensor is much lower than its pixel count.

By the way, there is nothing subjective about all of this. Likewise, there is no theoretical reason why the "colour" (colour saturation?) should be better with a Foveon sensor rather than a Bayer sensor. If the Sigmas produce better colour saturation it is because of a better implementation, not the Foveon versus Bayer technique.

PS. Sorry for contributing to turning your thread into a geekfest, Windswept!

… engrossed in such factional acts as dreaming different dreams.
  quote