The Joy of Pixel Density

Started Jul 13, 2008 | Discussions
Shop cameras & lenses ▾
John Sheehy Forum Pro • Posts: 18,566
The Joy of Pixel Density

Well, here is a preliminary example of what I have been promising for a little while; a direct comparison of pixel density with the same subject, same distance, same f-stop, same exposure time, same focal length. The cameras are the Canon 400D with 5.7 micron pixels, and the Panasonic FZ50 with 1.97 micron pixels. The larger crops are at the scale of 100% pixel view of the FZ50, and the smaller ones, at the scale of 100% pixel view of the 400D.

I say preliminary because this was taken from a distance, over a busy Manhattan highway, so there is a bit of atmospheric haze with pollution, limiting the sharpness of the FZ50 image, and they were hand-held, too. The G9 would also be good to include in a more thorough comparison, as it has better RAW data and less pattern noise than the FZ50, despite having slightly higher read noise as a whole.

These are RAW (simply interpolated to full RGB), with no WB, and no noise reduction, and no explicit sharpening. This is just to compare RAW noise per unit of area. As you can see, the unbalanced colors are almost exactly the same, making the comparison simpler than if the color channel responses were greatly different.

These are pushed to a true exposure index of ISO 13,500 from ISO 100 on both cameras.

-- hide signature --

John

aclo Regular Member • Posts: 480
Re: The Joy of Pixel Density

John Sheehy wrote:

Well, here is a preliminary example of what I have been promising for
a little while; a direct comparison of pixel density with the same
subject, same distance, same f-stop, same exposure time, same focal
length. The cameras are the Canon 400D with 5.7 micron pixels, and
the Panasonic FZ50 with 1.97 micron pixels. The larger crops are at
the scale of 100% pixel view of the FZ50, and the smaller ones, at
the scale of 100% pixel view of the 400D.

I just noticed all the threads going on recently about this pixel size business (I'm newish here, spent more time on usenet until a few months ago). Remembering the "fun" that was had by all the few times this was discussed on usenet, I am looking forward to the responses here

[nicely done by the way, but I'm willing to bet it won't convince those who don't want to be convinced-my guess is it'll take some time until a few "authorities" are convinced, and then slowly more people will wake up]

John Sheehy OP Forum Pro • Posts: 18,566
Re: The Joy of Pixel Density

8 hours, and only one reply. The silence is deafening. If I had started a thread about how bad the megapixel race is, the thread would be moving towards the 150 mark already. I guess people don't know what to do with reality when it contradicts their beliefs.

-- hide signature --

John

Graystar Veteran Member • Posts: 8,373
Re: The Joy of Pixel Density

John Sheehy wrote:

8 hours, and only one reply. The silence is deafening.

Sorry...this is the first time I see it.

John Sheehy wrote:

Well, here is a preliminary example of what I have been promising for
a little while; a direct comparison of pixel density with the same
subject, same distance, same f-stop, same exposure time, same focal
length. The cameras are the Canon 400D with 5.7 micron pixels, and
the Panasonic FZ50 with 1.97 micron pixels. The larger crops are at
the scale of 100% pixel view of the FZ50, and the smaller ones, at
the scale of 100% pixel view of the 400D.

I swear I always have such a hard time figuring out what you're saying! Did the same image fall on the same amount of area on both sensors? If focal length and distance are the same then it should, but you're showing two images that are the same size. The original images are not the same resolution. So when you show two images that are the same resolution and say that they're straight from the RAW...that confuses me. The FZ50 should have a higher resolution. Did you resize?

And what is this "pushing" that you're doing? Can't you display the demosaiced RAW without messing with the EC?

I just want two images...from the same sensor area...resize the higher resolution to match the lower resolution...of a scene with a wide dynamic range. That's it. Oh...maybe some clear labeling. And no pushing. Is that too much to ask?

Oh...and DNGs posted somewhere would also be nice

Barry Fitzgerald Forum Pro • Posts: 29,888
Re: The Joy of Pixel Density

John Sheehy wrote:

8 hours, and only one reply. The silence is deafening. If I had
started a thread about how bad the megapixel race is, the thread
would be moving towards the 150 mark already. I guess people don't
know what to do with reality when it contradicts their beliefs.

John, I am almost certain you are some implant from a major camera manufacturer, trying to justify more megapixels

Is this another one of those crop the APS SLR sensor size, to that of the FZ-50, in some odd attempt to prove that the panasonic has a better sensor?

-- hide signature --

I am not the 'Ghost Hunter', nor am I the Irish actor in the 'Quiet Man'

Wayne Larmon Veteran Member • Posts: 9,403
Nailing jelly to a tree.

John Sheehy wrote:

8 hours, and only one reply. The silence is deafening. If I had
started a thread about how bad the megapixel race is, the thread
would be moving towards the 150 mark already. I guess people don't
know what to do with reality when it contradicts their beliefs.

John, I agree with you, but the problem is that the story is intermingled amongst the 1,000+ posts about pixel density that have appeared over the course of the past week or so, in more threads than I can remember. I've read most of the posts in these threads, but, for the life of me, I can't remember where the exact posts of yours are that explained what you are proving in this thread.

It would help a whole lot if you could put up a web page that covered all the issues, along with the tests you've done that support your thesis. Forum posts are transitory. Trying explain a subject as complex as this is with multiple, fragmentory, forum posts is like trying to nail jelly to a tree.

It would be very helpful if the whole thing was in one place that we could refer to. Without needing to save a blizzard of bookmarks to individual forum posts.

Wayne

alanr0 Senior Member • Posts: 1,780
Re: The Joy of Pixel Density

John Sheehy wrote:

8 hours, and only one reply. The silence is deafening.

Well some of us are about to go to bed, and can't understand what's so difficult to understand in any case.

On the other hand ...

Graystar wrote:

I just want two images...from the same sensor area...resize the
higher resolution to match the lower resolution...

Which John has put in the two small inserts

of a scene with a
wide dynamic range. That's it. Oh...maybe some clear labeling. And
no pushing. Is that too much to ask?

With no pushing JPEG images won't show a wide dynamic range. That's the way JPEGs are. The images John has posted would be mostly black.

If John uses something close to the SRGB tone curve, anything not blown out in the posted image will have less than 8% pixel values (21/255) - somewhere between X and Y in the gray scale at the bottom of Phil's review pages. With the same 7 stop push, gamma=0.45 would put the brightest parts of the crop around 11% (X on the gray scale).

http://en.wikipedia.org/wiki/SRGB
http://www.w3.org/Graphics/Color/sRGB.html
With a linear tone curve, the image would be indistinguishable from black.

Are you still confused about the difference between dynamic range and tone curve?

Cheers.
--
Alan Robinson

Graystar Veteran Member • Posts: 8,373
Re: The Joy of Pixel Density

alanr0 wrote:

Graystar wrote:

I just want two images...from the same sensor area...resize the
higher resolution to match the lower resolution...

Which John has put in the two small inserts

And the small inserts are the same size, which they shouldn't be, so I don't know what they are.

Can't we just get the originals?

With no pushing JPEG images won't show a wide dynamic range. That's
the way JPEGs are. The images John has posted would be mostly black.

Are you saying that if the images were displayed in some other form...say printed...that the pushing wouldn't be necessary and the image wouldn't be mostly black?

John Sheehy OP Forum Pro • Posts: 18,566
Re: The Joy of Pixel Density

Graystar wrote:

John Sheehy wrote:

8 hours, and only one reply. The silence is deafening.

Sorry...this is the first time I see it.

John Sheehy wrote:

Well, here is a preliminary example of what I have been promising for
a little while; a direct comparison of pixel density with the same
subject, same distance, same f-stop, same exposure time, same focal
length. The cameras are the Canon 400D with 5.7 micron pixels, and
the Panasonic FZ50 with 1.97 micron pixels. The larger crops are at
the scale of 100% pixel view of the FZ50, and the smaller ones, at
the scale of 100% pixel view of the 400D.

I swear I always have such a hard time figuring out what you're
saying! Did the same image fall on the same amount of area on both
sensors?

Yes, of course. That is the best way to compare pixel density.

If focal length and distance are the same then it should,
but you're showing two images that are the same size.

I'm showing two crops where the subject is the same size in both (just as it was on the sensors). The difference is that one had 1.97 micron pixels (higher pixel density), and the other had 5.7 micron pixels (lower pixel density). The higher pixel density gives both better resolution and less practical noise in these poorly lit shadows.

The original
images are not the same resolution.

Both cameras are 10MP cameras.

So when you show two images that
are the same resolution and say that they're straight from the
RAW...that confuses me. The FZ50 should have a higher resolution.
Did you resize?

Of course I did. I explained that quite clearly in the first post.

And what is this "pushing" that you're doing? Can't you display the
demosaiced RAW without messing with the EC?

Why? They would just be dark images with only light bulbs visible. That would tell us nothing about how sensitive the sensors are in the deep shadows vs noise.

I just want two images...from the same sensor area...resize the
higher resolution to match the lower resolution...of a scene with a
wide dynamic range.

This one isn't DR, per se. This is about absolute noise, as a function of pixel noise and displayed pixel size. You are imagining that a pure, accurate DR test is an easy thing to do. You have to match lighting to RAW saturation, to have the exact same target appear exactly the same number of stops away from RAW saturation. Lots of trial and error.

That's it. Oh...maybe some clear labeling.

Like "400D", and "FZ50"?

And
no pushing. Is that too much to ask?

Yes it is, because I can't trust people in general to figure out how to do it themselves.

-- hide signature --

John

John Sheehy OP Forum Pro • Posts: 18,566
Re: The Joy of Pixel Density

Barry Fitzgerald wrote:

John Sheehy wrote:

8 hours, and only one reply. The silence is deafening. If I had
started a thread about how bad the megapixel race is, the thread
would be moving towards the 150 mark already. I guess people don't
know what to do with reality when it contradicts their beliefs.

John, I am almost certain you are some implant from a major camera
manufacturer, trying to justify more megapixels

That's what you have to tell yourself. I am only interested in the truth. I could never market something I didn't truly believe in.

Is this another one of those crop the APS SLR sensor size, to that of
the FZ-50, in some odd attempt to prove that the panasonic has a
better sensor?

No. It is to prove that the sensor is better per unit of area than DSLR sensors, because it has a higher pixel density, which is just a fact, that some people seem to have problems dealing with.

-- hide signature --

John

Barry Fitzgerald Forum Pro • Posts: 29,888
Re: The Joy of Pixel Density

John Sheehy wrote:

No. It is to prove that the sensor is better per unit of area than
DSLR sensors, because it has a higher pixel density, which is just a
fact, that some people seem to have problems dealing with.

But if you take a crop from the APS sensor, to make it so low resolution, its obviously not going to have enough res to make up a decent image. You know the numbers, so what res is the APS image??

Had you showed it on the same sized sensor, with various levels of mp's, and compared, that might make sense.

-- hide signature --

I am not the 'Ghost Hunter', nor am I the Irish actor in the 'Quiet Man'

John Sheehy OP Forum Pro • Posts: 18,566
Re: The Joy of Pixel Density

Barry Fitzgerald wrote:

John Sheehy wrote:

No. It is to prove that the sensor is better per unit of area than
DSLR sensors, because it has a higher pixel density, which is just a
fact, that some people seem to have problems dealing with.

But if you take a crop from the APS sensor, to make it so low
resolution, its obviously not going to have enough res to make up a
decent image.

That's part of the point! I am not giving the big pixels a handicap; that's the way they ARE . They are blunt, and unresolving. How many of them there are in the full image is totally irrelevant to the issue of pixel density.

You know the numbers, so what res is the APS image??

Had you showed it on the same sized sensor, with various levels of
mp's, and compared, that might make sense.

That would make sense, TOO. However, it does not exist, so that is why it must be done this way.

-- hide signature --

John

Graystar Veteran Member • Posts: 8,373
Re: The Joy of Pixel Density

John Sheehy wrote:

Of course I did. I explained that quite clearly in the first post.

Clearly...right. Anyways I think I got it. On the left side the inset image is the 100% crop from the 400D and the large image is an upsizing of that crop...while on the right side the large image is the 100% crop from the FZ50 and the inset is the downsized image of that crop. Is that it?

Why? They would just be dark images with only light bulbs visible.
That would tell us nothing about how sensitive the sensors are in the
deep shadows vs noise.

Why? Because I only care about shadow detail that I can see. Yes, you reduced the noise level so much that Imatest can clearly distinguish bands of pixels with values of 1, 2, and 4...wow! two more stops! Except that I can’t see it or print it so it’s useless.

And no pushing. Is that too much to ask?

Yes it is, because I can't trust people in general to figure out how
to do it themselves.

Is that why you didn’t post any DNGs?

Jay Turberville Forum Pro • Posts: 12,917
Another example

Hey John, I just did some processing of image today and posted the results to last weeks thread. I'll post them here as well.

The comparison camera is an E-300 8Mp 4/3" sensor equipped DSLR and a Coolpix 8400 with an 8Mp 2/3" sensor. Both cameras were introduced at almost the same exact time. The assumption is that they have similar levels of technological developement. I downsampled two images from the 8400. The idea is to show what happens when you software "bin" the image info from a sensor with twice the pixel density.

The first was DPReviews resolution test image. Even when a gaussian blur and a 50% resize is applied, the resolution does not actually reduce to half resolution. The second was one of my step wedge images that tests for dynamic range.

Here's the DR of the E-300.

This image shows how the usable DR improves upon downsizing and exceeds the E-300.

This image demonstrates that not only is the DR improved, but the resolution is not halved. So the greater pixel density give you better DR and more detail. Though I should add that it does so at a lower ISO (50 compared to 100).

-- hide signature --
ejmartin Veteran Member • Posts: 6,274
Re: The Joy of Pixel Density

John Sheehy wrote:

No. It is to prove that the sensor is better per unit of area than
DSLR sensors, because it has a higher pixel density, which is just a
fact, that some people seem to have problems dealing with.

Are you sure it's the higher pixel density? Or is it that the smaller DR of the smaller pixels puts less demands on the DR of the VGA/ADC, so that the small pixel camera is limited by the pixel DR while the large pixel DSLR sensor is limited by the VGA/ADC and can't demonstrate its full potential? It's not obvious to me that you've proven that the small pixel sensor is better per unit area; rather you have shown that the sensor/VGA/ADC combination favors the small pixel sensor. One still has to disentangle which component of that combination is responsible.

Sorry to play the devil's advocate here.

-- hide signature --
Jay Turberville Forum Pro • Posts: 12,917
Yet another example

In this example, I make a direct comparison between two different cameras rather than extrapolate how a 32Mp 4/3" sensor's performance from the performance of an 8Mp 2/3" sensor.

Here I shot the same subject with the camera's at the same ISO (one stop greater than base ISO for the 8400), nearly the same focal length (21mm) and similar exposures. The exposures are only similar because the E-300 image was noticably darker than the CP8400 when the same exposures were used. The E-300 image received about 1/3 stop more light than the CP8400 image.

So, the two optical images were formed at nearly identical actual sizes on the sensors. The difference is that the 8400 sensor is smaller and records less total angle of view. To put the two images from on equal footing for comparison, I downrezzed the CP8400 image. This optimizes the 8400 image for noise and dynamic range at the expense of resolution. I then boosted the shadows a bit in Photoshop.

The DSCN image is the Coolpix image. It shows more detail. But does it have more shadow detail? Well I boosted the shadows even more for each image by the same exact amount in Photoshop and the images below are the result. (Note that they are in reverse order). The Coolpix image (DSCN) shows slightly more shadow detail, more detail overal, and noticably less chroma noise. The E-300 image is marred by a lens with more CA though. Also, keep in mind that the E-300 image received one-third stop more exposure.

If the Coolpix 8400 were shot at ISO 50, it would have a clear edge in DR. Also, keep in mind that we traded image detail for DR in the 8400 image. If we didn't want or need the extra DR, we could get far more detail with the 8400 image, which is showing us what we'd expect from 32Mp 4/3' sensor - given the optics were up to the task.

-- hide signature --
Jay Turberville Forum Pro • Posts: 12,917
Re: Another example

In case it wasn't clear (I didn't word that post very well), the point is that if you crammed 32 million of the CP8400 pixels onto a 4/3" sensor camera, you could process the image so that you end up with both resolution and dynamic range that is superior to what you get with the E-300 camera. The downrezzing is to software "bin" the pixels to get an image resolution that is approximately the same resolution as you get with an 8Mp 4/3" sensor. In fact, the resolution is a bit greater and shows less aliasing. So overall, this shows that a 32Mp 4/3" sensor can compete quite well with an 8Mp 4/3" sensor in the area of usable DR.

-- hide signature --
Rriley
Rriley Forum Pro • Posts: 21,846
Re: Another example

i know you would have thought about this Jay, what would the impact of diffraction be on a hypothetical 32Mp 4/3rds sensor/lens system. Given the following article from LL.

http://www.luminous-landscape.com/tutorials/resolution.shtml

and knowing you can do a lot with just (just) F4.5 for deep dof on 4/3rds, like this 11mm F 4.5 at iso 800 on E3, where its a full 60ft to that furthest window

not noticeable here but it is a bit noisy, by experience even when shots like this are bill boarded to large sizes the noise isnt apparent. Still i dont think i would have done better with FF at iso3200

-- hide signature --

Riley

in my home, the smoke alarm is the dinner bell (just)

 Rriley's gear list:Rriley's gear list
Sigma DP2 Merrill Canon EOS 5D Olympus E-3 Olympus E-5 Panasonic Lumix DMC-GH2 +1 more
ZorSy Veteran Member • Posts: 3,480
Re: The Joy of Pixel Density

John,

I re-read this post several times and I am intrigued with what you are saying - sorry if I am slow to understand (it's not the language but method) so please correct me if I am wrong here:

you are comparing pixel density per surface units, right? SO for your test you actually used as much surface from 400D sensor as much is the FZ50 CCD surface, resulting in fewer total pixel count from 400D, for which reason you had to uprez fewer pixels from that sample to match the total number of "true" pixels full FZ sensor provided? In my understanding of presented, you are comparing the uprezzed crop (which is now 100%) to a 100% crop from FZ (and then the whole lot being pushed up)?

If my understanding is not right, could you please explain again your workflow for this test. In your OP you said the shooting parameters were the same including the focal lenght, I guess equaling it xx/35mm. Yet this is where the term "area" comes in place before the pixel count.

I did something similar comparing two different cameras (D80 and Fuji 9500, the second using SCCD resulting in 18MP interpolated JPEG), but my findings are somewhat quite opposite to yours, even using base ISO 80 on Fuji.

cheers

 ZorSy's gear list:ZorSy's gear list
Nikon D80 Nikon D7100
Jay Turberville Forum Pro • Posts: 12,917
Re: Another example

I'm not going to read the article tonight. Frankly, that site doesn't usually inspire with their good analysis. But here's my short answer. If the sensor outresolves the lens, you lose nothing. You can always downsample to match the capabilities of the lens, and as I demonstrated, you are not likely to lose any usable DR due to the smaller pixel pitch. Further, you can afford to have a less agressive AA filter with the higher pixel pitch.

-- hide signature --
Keyboard shortcuts:
FForum MMy threads