Too many pixels for hand-held?

[pico]

Advertisement (gone after registration)

Let's seem some real world examples. Until then I call BS.

[jaapv]

Calling BS does nothing to repeal simple geometry. So feel free.

[jaapv]

I see an excellent likeness of Homer Simpson.

 

What has always puzzled me about "pixel blur" assertions is that the problem should be most pronounced in cell phone cameras.

They have many pixels crammed into miniature sensors.

So how come we don't see complaints from users of cell phone cameras, but we do see concerns from users of high-end cameras?

Maybe a different standard of sharpness? The screen makes for thumbnail-sized images, so just about everything will appear sharp anyway.

[harmen]

The physical size of the sensor should not matter in this theory.  The image gets projected onto the full sensor no matter what size the sensor is.  If one angles or shakes the camera by a degree, the number of sensels that the image shifts across would always be equal.

 

The more interesting part, mentioned by someone earlier, is whether the motion that is captured more clearly by having more sensels shows up when printing at equal size to an image taken by a sensor with fewer sensels.  Imagine a sensor with 100 times the number of sensels.  It would have 10x more horizontal and 10x more vertical sensels.  If with a certain amount of camera shake, motion would have caused a half sensel shift in the small count sensor, it would cause shift across 5 sensels in the higher mp sensor.  However, when printing, does that actually matter?  It would when zooming in to pixel level, but that's different.

 

btw, I'm not sure that my photos gain from more pixels that I don't print (or display), especially when sharpness is not what they are all about.

[zlatkob]

It is not a myth. See  my slightly primitive sketch.

 

On a large pixel the motion blur is inside the sensel, thus invisible.

With small pixels the same amount of blur will affect three sensels, making it visible and even artificially increasing it.

 

So the problem is not the MP count, but the pixel size. A medium format 50 MP sensor will have less problems with motion blur than a 35  50 MP sensor.

 

pixel.jpg

 

Obviously, the larger the blur gets (thus outside the large sensel), the less the impact of the spillover -relatively- will be.

 

 

The degree of motion is exactly the same in those examples.  You could only every spot a difference if you look at individual pixels, which is an extreme level of magnification.  In other words, you are no longer looking at a photograph, but instead looking at a microscopically small portion of a photograph. Changes that occur at that level are not evident in any regular print size (for me).  I've never made a print so large that I could see individual pixels in it.  If there were a difference, it would be extremely trivial.  Perhaps there are people making such large prints that a change in one pixel matters for them.

[edwardkaraa]

One counter argument is that with more pixels, there is more data available to the downsizing algorithms to work with, resulting in slightly better overall quality to a given output/print size. A few guys did some tests in a thread on FM and shown this to be the case. Though I have to admit that I found the quality improvement so difficult to distinguish it makes really no difference in practical use.

[marchyman]

Advertisement (gone after registration)

 

So how come we don't see complaints from users of cell phone cameras, but we do see concerns from users of high-end cameras?

 

 

I saw homer, too

 

Many cell phone camera users look at pictures on their cell phones.  Slightly out of focus and/or motion blurred images somehow don't look so bad on a 4" screen.  That, and many don't care.

 

I have a hard time taking a cell phone image without motion blur.  Sometimes I think I've done OK -- until I look at the image on my 27" monitor.

[Exodies]

Have you seen the Apple advertising campaign, large posters of people's iPhone pictures? Not limited by blur or close observation.

[jaapv]

The degree of motion is exactly the same in those examples.  You could only every spot a difference if you look at individual pixels, which is an extreme level of magnification.  In other words, you are no longer looking at a photograph, but instead looking at a microscopically small portion of a photograph. Changes that occur at that level are not evident in any regular print size (for me).  I've never made a print so large that I could see individual pixels in it.  If there were a difference, it would be extremely trivial.  Perhaps there are people making such large prints that a change in one pixel matters for them.

Of course the amount of blur is the same. That is exactly the point. In example 1 the camera records it, in example 2 the camera records even more than is present.

 

It does make a difference if the blur is contained within one pixel or is recorded by three pixels. (in reality the argument goes for groups of pixels, as they are interpolated by the Bayer demosaicing algorithm) You lose edge contrast, which translates as sharpness. What you see in one pixel is added millions of times when looking at the photograph, so these small differences do make a difference.

If it were otherwise it would not make a difference whether you use a low resolution or high resolution sensor unless you print to pixel size.

 

One can't have it both ways; first claiming high resolution sensors make for technically better images and then arguing it does not matter what is recorded on pixel level.

 

By this argument LP/mm doesn't matter on film because it is on a microscopic level, aspherical lenses do not matter because they are measured in nanometers, etc.

[zlatkob]

It does make a difference if the blur is contained within one pixel or is recorded by three pixels. (in reality the argument goes for groups of pixels, as they are interpolated by the Bayer demosaicing algorithm) You lose edge contrast, which translates as sharpness. What you see in one pixel is added millions of times when looking at the photograph, so these small differences do make a difference.

If it were otherwise it would not make a difference whether you use a low resolution or high resolution sensor unless you print to pixel size.

One can't have it both ways; first claiming high resolution sensors make for technically better images and then arguing it does not matter what is recorded on pixel level.

 

A high resolution sensor makes a technically better image when it records more detail and the resulting image is shown large enough for the eye to see that detail.  But if it records motion blur, the size of the blur within the image is the same regardless of the resolution.  While a higher resolution sensor can make the sharp detail in an image sharper, it won't make the motion blur sharper — at least not in any photographically significant way.  One would have to enlarge enough to count individual pixels to see any describable difference in the blur.  I think we are talking about microscopic differences, not differences that affect normal photography.

 

Over the years, the megapixels of my cameras have increased at least five times, but their hand-holdability hasn't changed a bit.  How can that be?  Another doubling in megapixels won't make a difference.  I still use the same shutter speeds and still make the same-sized prints.  I still don't consider 100%-pixel-viewing as my output.  Further, as resolution continues to increase, the likelihood of my ever considering 100%-pixel-viewing as my final output continues to decrease, not increase.

 

Anyway, it's not something we can resolve with thought experiments.  This is truly a question that can be resolved with real world examples.  Presumably someone out there who has claimed that higher resolution makes it more difficult to do hand-held photos has shown this with actual comparison photos shown at the same output size (not 100% pixel viewing).  I've never seen it.

[jaapv]

Well, that probably is more telling about your handholding technique than about the technical discussion

 

I'm a bit lost by your remark about making blur sharper - Blur is made blurrier, that is the argument.

 

Just one thing: Photography is about the sum of microscopic differences, so don't discount it. You are working with light, which, in itself, is nanometer-small.

[ralphh]

It is not a myth. See  my slightly primitive sketch.

 

On a large pixel the motion blur is inside the sensel, thus invisible.

With small pixels the same amount of blur will affect three sensels, making it visible and even artificially increasing it.

 

So the problem is not the MP count, but the pixel size. A medium format 50 MP sensor will have less problems with motion blur than a 35  50 MP sensor.

 

pixel.jpg

 

Obviously, the larger the blur gets (thus outside the large sensel), the less the impact of the spillover -relatively- will be.

 

Your diagram assumes your eyes can actually see individual pixels. I would suggest that at print sized below about 50 cm you probably cannot actually do so, which comes back to my point about looking at images on a PC screen at 100% vs a print. 

 

Also, you say the blur is inside a pixel, hence invisible,.. Well a pixel can only be one color. By definition the entire pixel is a perfect blur, regardless of what it's showing, so in the top you have a perfect blur the width of 3 smaller pixels. Unless those three pixels are all so blurred as to show exactly the same colour, they will have more detail and appear less blurred than the single pixel

 

lets imagine a hard black/white edge running vertically down your diagram. In the top, your pixel will be perfect 50% grey, ie blurred for its entire width. With the 3 pixels, assuming at the one on one side ends up a little more white, and the other side a little more black, it'll look sharper at the some output size.

 

on the screen at 100% where all pixels are the same size, the higher res will look blurrier as you 3 pixels of blur not one, but the you've effectively enlarged 3 times more

[jaapv]

The problem is that the print you see is composed of individual pixels, as evidenced by for instance Monochrom prints which, liberated from demosaIcing, do show better acuity as the pixels are transferred 1:1.

[ralphh]

But you are never printing 1:1. You are generally printing at a given print size. 

 

Yes if if you print 1:1 less pixels will look sharper, ie if you make a 2m print from 24mp and 3m print from 50mp and look at them from the same distance, the 24 mp one will look sharper. 

 

If if you want to prove it one way or the other, try shading in your pixels in your diagram and show us an example where the one pixel has more detail than the 3

 

if you take your argument to its logical conclusion, a 1000MP camera would be unusable, as where a 1 pixel (not 1mp, one pixel) would give the ultimate in sharpness

[jaapv]

Taken to extremes, that is correct, although it would mean a two pixel camera to create an edge.

[ralphh]

But not a black white edge- you'd have one dark grey or one light grey pixel, depending on where the edge fell, so your blur is still the entire widthof the picture.

[Peter H]

To check that I'm understanding this may I ask this: If the print size is constant when comparing two sensors, won't the fact that the increased blur of the sensor with more pixels be outweighed ( or at least counterbalanced) by the fact that each occurrence of blur will be smaller when compressed into the same sized print?

[jaapv]

Whatever, this is of course a wholly unrealistic situation, rather reminiscent of  LP/mm - contrast.

[jaapv]

To check that I'm understanding this may I ask this: If the print size is constant when comparing two sensors, won't the fact that the increased blur of the sensor with more pixels be outweighed ( or at least counterbalanced) by the fact that each occurrence of blur will be smaller when compressed into the same sized print?

Yes. But using larger pixels on an MF sensor will yield a better result at the same MP count.

[Advertisement]

Hallo,

Look here: Leica M

[Exodies]

That "counterbalanced" is the crux of the biscuit. It's very difficult to theorize about digitised phenomena. Someone should do the maths.