Effective Pixels in Digital Cameras

clay@claysturner.com wrote:
> On Thursday, November 28, 2013 8:44:45 AM UTC-5, Greg Berchin wrote:

(snip)
>> So am I right in interpreting that 6000x4000 Bayer-Filtered image as consisting
>> of 6 million REAL pixels, or have I misinterpreted something?
 
> You are correct that there are no sub pixels with Bayer type sensors. For example 
> a 12MP sensor will have 6MP green, 3MP red, and 3MP blue pixels. When the image is 
> demosaiced (deBayered) you will end up with 12M samples where each sample has 
> an R,G, and B component.

An interesting side effect of this process is the ability to detect
modifications to the image. If you splice parts of images together, the
result likely won't be the result of demosaicing of any set of sensor
data.
 
> Obviously aliasing becomes a concern but is not a often a big problem. Since the 
> lens itself functions as a low pass filter in terms of spatial frequencies, the 
> sensor doesn't need much of an anti-alias (low pass) filter built in.

Less true in the early days of digital photography, but pretty true now.
 
> Yes, high sensor densities demand large apertures (small F numbers) if you 
> wish to not blur the image at the pixel level.
 
> In terms of 35mm film, the diffusion of the color dyes often limited the 
> maximum spatial resolution. Popular Photography did a resolution analysis 

Usually called dye clouds, as there aren't any grains left in a color
image. When the AgBr grain is developed, the oxidized color developer
difuses to a nearby color coupler to form a dye molecule. The result
doesn't look the same as the sharp silver grains of B&W photography.

> of hundreds of slides that were used for photos used in the magazine and 
> they found the highest observed resolution was about 60 line 
> pairs/millimeter (lp/mm). Most topped out in the 40 lp/mm range. 
> A 12MP Bayer sensor in a modern camera will far exceed 35mm film in 
> terms of linearity, resolution, and dynamic range. I converted to 
> digital over a decade ago and haven't looked back since.

There are published MTF graphs for many filems, but I haven't thought
about the number, or compared them to B&W films.

This does remind me of something else, though, the change from line
pairs per (lenght unit) and dots per (length unit), and even more that
line pairs was traditionally, and still sometimes is, written at lines
per (length unit). That is, a factor of two change. 

-- glen

Greg Berchin <gjberchin@chatter.net.invalid> wrote:
> On Thu, 28 Nov 2013 20:36:49 -0800 (PST), clay@claysturner.com wrote:
 
>> In terms of 35mm film, the diffusion of the color dyes often 
>> limited the maximum spatial resolution. Popular Photography did 
>> a resolution analysis of hundreds of slides that were used for 
>> photos used in the magazine and they found the highest observed 
>> resolution was about 60 line pairs/millimeter (lp/mm). 
>> Most topped out in the 40 lp/mm range. A 12MP Bayer sensor in 
>> a modern camera will far exceed 35mm film in terms of linearity, 
>> resolution, and dynamic range. I converted to digital over a 
>> decade ago and haven't looked back since.

But also, the resolution of analog systems has a fairly long tail,
as contrast decreases with increasing spatial frequency. Digital
images have a sharp limit, though also the filtering due to finite
sensor size. 
 
> OK, having now established that my suspicion was correct (that 
> the theoretical spatial resolution limit of an NxM digital camera 
> image sensor is [N/2]x[M/2]), let us move on to briefly examine 
> these other issues.
 
> As I mentioned earlier, I have a 4032x2688 35mm film scanner that 
> can not resolve grain in fine-grain B&W film. Tom Gardner has stated 
> that he needed 5000x3000 to resolve grain in color slides. 
> I vaguely remember photo magazine tests from 30 years ago only 
> occasionally stating that lens resolution was "diffraction 
> limited" at small apertures (F32 and higher numbers).

I was once scanning old (from my grandfather) Tri-X and believe that
I did resolve the grain.  But maybe not with new Tri-X or for
Panatomic-X of any age. 

My old favorite from years ago was Panatomic-X in Diafine at ISO 250.

The usual film scanners use either a three line sensor array, one for
each color, or an unfiltered array and three LEDs as light sources.
So, no Bayer array.

Also, note that there are a variety of different filter arrays, many
of which allow more than one of (R, G, B) through some or all of the
filters. Some use (R, G, B, W) as the four in a square, with W being
white (no filter). Others combinations with Cyan, Yellow and Magneta,
each allowing two of (R, G, B). 

But as far as the original question, if the image contains alternating 
black and white stripes, as the usual resolution targets do, you have
the full, not half, resolution.

-- glen

On 29/11/13 13:51, Greg Berchin wrote:
> As I mentioned earlier, I have a 4032x2688 35mm film scanner that can not
> resolve grain in fine-grain B&W film. Tom Gardner has stated that he needed
> 5000x3000 to resolve grain in color slides.

5000*3000 did /not/ resolve the grain.

Any higher would have resolved the grain, but unfortunately
after 15 years I don't recall what "any higher" meant.

On Friday, November 29, 2013 5:18:58 AM UTC-8, Greg Berchin wrote:
> ... 
> I was comparing the pixel configuration to that of monitors, and from that
> implicitly deriving the resolution capability from the pixel count. My shocking
> revelation (shocking to me, anyway) was that, unlike the resolution limit of a
> monitor, which is related to its pixel spacing, the resolution limit for a
> digital camera image sensor is related to _twice_ its pixel spacing.
> (Conventional underlining used in deference to dbd.)
> ...

I stated no objection to your choice of method to emphasize content. But -you- seem to have been sensitive to the common name of the use of all caps for emphasis. My point was that the content you chose to emphasize was irrelevant 
and incompetent with regard to determining the resolution of a camera or display as is any undefined act of "implicitly deriving the resolution capability from the pixel count".

> ...

Dale B. Dalrymple

On Friday, November 29, 2013 5:51:49 AM UTC-8, Greg Berchin wrote:
> On Thu, 28 Nov 2013 20:36:49 -0800 (PST), clay@claysturner.com wrote:
> 
> >In terms of 35mm film, the diffusion of the color dyes often limited the maximum
> >spatial resolution. Popular Photography did a resolution analysis of hundreds of
> >slides that were used for photos used in the magazine and they found the highest
> >observed resolution was about 60 line pairs/millimeter (lp/mm). Most topped out
> >in the 40 lp/mm range. A 12MP Bayer sensor in a modern camera will far exceed
> >35mm film in terms of linearity, resolution, and dynamic range. I converted to
> >digital over a decade ago and haven't looked back since.
> 
> Thanks, Clay.
> 
> OK, having now established that my suspicion was correct (that the theoretical
> spatial resolution limit of an NxM digital camera image sensor is [N/2]x[M/2]),
> let us move on to briefly examine these other issues.

I don't believe that anything posted in this thread so far serves to justify that suspicion. If you chose to study the issue look for actual measurements based on specifications of resolution such as MTF (modulation transfer functions) in places such as:
http://www.imatest.com/docs/sharpness/
and the others Clay has suggested.

> ...

Dale B. Dalrymple

On Fri, 29 Nov 2013 16:31:40 -0800 (PST), dbd <dbd%ieee.org@gtempaccount.com>
wrote:

>I don't believe that anything posted in this thread so far serves to justify 
>that suspicion. If you chose to study the issue look for actual measurements 
>based on specifications of resolution such as MTF (modulation transfer 
>functions) in places such as: http://www.imatest.com/docs/sharpness/
>and the others Clay has suggested.

Thank you, Dale. 

Dale, in other threads your responses have been consistently insightful and
helpful. In this one, you have been insulting and, up to now, not helpful. The
link that you provided, however, is helpful.

As for justification of my suspicion, I quote from _your_ citation:

"Sensor response above the Nyquist frequency is garbage. It can cause aliasing,
visible as Moire patterns of low spatial frequency ..."

and

"[Nyquist Frequency] The highest spatial frequency where a digital sensor can
capture real information. Nyquist frequency fN = 1/(2 * pixel spacing) = 0.5
cycles/pixel."

The latter statement is the fundamental basis for this whole thread, and shows
just how "incompetent" it is to determine the limits on spatial resolution of an
imaging sensor from the spacing of the pixels. 

There are many things that can reduce the achieved spatial resolution relative
to the theoretical. For this thread, I don't care about them.

Greg

On Saturday, November 30, 2013 4:54:56 AM UTC-8, Greg Berchin wrote:
> ...
> Dale, in other threads your responses have been consistently insightful and
> helpful. In this one, you have been insulting and, up to now, not helpful. The
> link that you provided, however, is helpful.

When people come to comp.dsp to ask questions and learn I am happy if I can help. You started this thread with claims of experience and proceeded to claims misusing signal processing principles.

> 
> As for justification of my suspicion, I quote from _your_ citation:
> "Sensor response above the Nyquist frequency is garbage. It can cause aliasing,
> visible as Moire patterns of low spatial frequency ..."
> 
> and
> 
> "[Nyquist Frequency] The highest spatial frequency where a digital sensor can
> capture real information. Nyquist frequency fN = 1/(2 * pixel spacing) = 0.5
> cycles/pixel."
> 
> The latter statement is the fundamental basis for this whole thread, and shows
> just how "incompetent" it is to determine the limits on spatial resolution of an
> imaging sensor from the spacing of the pixels.

Yes, this is my point exactly. The Nyquist frequency is an unreachable limit, and yet you persist in suggesting the unreachable as a default assumption. I consider that unsafe and inappropriate and have applied emphasis to my opinion.

> There are many things that can reduce the achieved spatial resolution relative
> to the theoretical. For this thread, I don't care about them.
> ...

This is a curious turnabout for someone who began the thread questioning the meaning of REAL, 'real' and 'effective' pixels. Their significance depends on the digital camera and film resolution they actually represent and that resolution was what you claimed to care about.

Dale B. Dalrymple

Greg Berchin wrote:

> As I mentioned earlier, I have a 4032x2688 35mm film scanner that can not
> resolve grain in fine-grain B&W film. Tom Gardner has stated that he
> needed 5000x3000 to resolve grain in color slides. I vaguely remember
> photo magazine tests from 30 years ago only occasionally stating that lens
> resolution was "diffraction limited" at small apertures (F32 and higher
> numbers).

With a Nikon LS5000 (4000 ppi, a 36x24 slide has 20 Megapixels) one can also
see grain in color slides.

> This seems to indicate that typical digital camera image sensors are still
> nowhere near the resolution limits of film or lenses.

Have a look here, where 6x7 cm medium(!) format film is compared to
a 36x24mm digital camera (Canon EOS 1Ds) with 11 Megapixels:

  http://www.luminous-landscape.com/reviews/shootout.shtml

Now consider using a modern Nikon D800 with 36 Megapixels:

  http://kbesios.com/blog/2012/07/06/nikon-d800e-vs-6x9-medium-format-film

bye
Andreas
-- 
Andreas H�nnebeck | email: acmh@gmx.de
----- privat ---- | www  : http://www.huennebeck-online.de
Fax/Anrufbeantworter: 0721/151-284301
GPG-Key: http://www.huennebeck-online.de/public_keys/andreas.asc
PGP-Key: http://www.huennebeck-online.de/public_keys/pgp_andreas.asc

On Mon, 02 Dec 2013 18:19:52 +0100, Andreas Huennebeck <acmh@gmx.de> wrote:

>> This seems to indicate that typical digital camera image sensors are still
>> nowhere near the resolution limits of film or lenses.
>
>Have a look here, where 6x7 cm medium(!) format film is compared to
>a 36x24mm digital camera (Canon EOS 1Ds) with 11 Megapixels:
>
>  http://www.luminous-landscape.com/reviews/shootout.shtml
>
>Now consider using a modern Nikon D800 with 36 Megapixels:
>
>  http://kbesios.com/blog/2012/07/06/nikon-d800e-vs-6x9-medium-format-film
>

Wow. Thank you. That is very compelling.

Greg

On Tuesday, December 3, 2013 10:50:17 AM UTC-5, Greg Berchin wrote:
> On Mon, 02 Dec 2013 18:19:52 +0100, Andreas Huennebeck <acmh@gmx.de> wrote:
> 
> 
> 
> >> This seems to indicate that typical digital camera image sensors are still
> 
> >> nowhere near the resolution limits of film or lenses.
> 
> >
> 
> >Have a look here, where 6x7 cm medium(!) format film is compared to
> 
> >a 36x24mm digital camera (Canon EOS 1Ds) with 11 Megapixels:
> 
> >
> 
> >  http://www.luminous-landscape.com/reviews/shootout.shtml
> 
> >
> 
> >Now consider using a modern Nikon D800 with 36 Megapixels:
> 
> >
> 
> >  http://kbesios.com/blog/2012/07/06/nikon-d800e-vs-6x9-medium-format-film
> 
> >
> 
> 
> 
> Wow. Thank you. That is very compelling.
> 
> 
> 
> Greg

One thing to remember about grains is they are either on or off. It is the statistical combination of "ons" and "offs" that create an effective midlevel. So don't let the scanner resolution needed for resolving grains indicate a picture's resolution.

A real hint to the effective resolution limit comes from the maximum F number you can use before diffraction starts to harm the resolution.

The greater the pixel density on the sensor, the lower your maximum F number is before diffraction dominates. Of course the pixel density is a function of both the number of pixels and the sensor size.

With a 12MP (FX size (24x36mm) sensor) stopping beyond F/13 starts to lose resolution.

With the D800E (36MP FX sensor) stopping past F/8 starts to throw away resolution!

Most films were okay at F/16! My ancient 6MP D100 (DX sensor 16x24mm) was okay at F/16.

Clay