• Please use real names.

    Greetings to all who have registered to OPF and those guests taking a look around. Please use real names. Registrations with fictitious names will not be processed. REAL NAMES ONLY will be processed

    Firstname Lastname

    Register

    We are a courteous and supportive community. No need to hide behind an alia. If you have a genuine need for privacy/secrecy then let me know!
  • Welcome to the new site. Here's a thread about the update where you can post your feedback, ask questions or spot those nasty bugs!

The 12MP Canon Digital Rebel April 27th

Asher Kelman

OPF Owner/Editor-in-Chief


Click on Image to go to Amazon site!

Main Features of the digital Xsi: This 12MP camera like the 1DsIII and 1DIII uses the digic III processor in 14 BIT and has a 12.2 MP APS-C size CMOS Sensor.

Speed: It boasts 3.5 frames per second

Self cleaning: Of course

Advantages and payoffs for jump to 14 BIT processing path:
Auto-optimiser brings illumination to shadows and and highlight tonal priority as is the 1Dx3 flagship cameras.

Live View: With grid to keep things straight if you wish but alas new level function like the Ricoh GRDII reviewed by John Nevill in OPF here.

Asher
 
Damn, those entry level dslrs are now better than 5 years old top-shelf 1D-series bodies... I can only imagine what next generation is gonna be like:)
 

Asher Kelman

OPF Owner/Editor-in-Chief
Downside: unless we can harness new physics, we are still faced with small apertures rippling the light that "scrapes" the edge of the aperture and which will be resolved by the fine pitch of the fine sensels which now will span the waves from one point of light,

So beyond about f5.6, expect diffraction robbing of the image of some detail in the print viewed at normal distances. So be prepared to check on your final required magnification for these prints.

Asher
 

John Sheehy

New member
Downside: unless we can harness new physics, we are still faced with small apertures rippling the light that "scrapes" the edge of the aperture and which will be resolved by the fine pitch of the fine sensels which now will span the waves from one point of light,

So beyond about f5.6, expect diffraction robbing of the image of some detail in the print viewed at normal distances. So be prepared to check on your final required magnification for these prints.

As long as the shot-to-shot speed stays reasonable, I don't mind the extra storage requirements of the higher pixel density cameras, even if the pixel-to-pixel sharpness begins to drop (memory cards are almost free these days). Oversampling is wonderful; really. If you undersample and get high pixel contrast (like in a Sigma DSLR with a sharp lens), that "pop" that is so desired by many people completely disappears with any minor resampling, rotation, perspective correction, etc. A very over-sampled capture experiences no such difficulties.

I do think that most of the talk about "diffraction limits" is exaggerated, or misapplied, though. Higher pixel density never, ever makes diffraction worse! Just the opposite is really true. The path of a photon going through a lens and hitting a sensor hits a number of obstacles that make it go off-course. Diffraction has the same effect, regardless of pixel pitch; the pixel pitch is totally irrelevant at the point that diffraction occurs. When the photon hits the sensor, then, the larger the pixels, the further off the photons can be, because a larger area of sensor surface is confused in a monolithic pixel. Big pixels are a source of blur, just like diffraction is. If it seems that big pixels are less affected by diffraction, that's only because the big pixels are causing far more blur than the diffraction is!
 

Asher Kelman

OPF Owner/Editor-in-Chief
John, You made me laugh! Yes that's true, one could have, after all one huge pixel and that may well degrade the image to oblivion. Still, for a given pixel size, one shouldn't ignore the aperture. Look at the 100% cut out of the image of the flower here. The aperture was f32 for a ~6.8 micron pitch sensel. See how the green has been bled into the edge of the white petal. Well you might argue that the petal does indeed have a green edge, but I wouldn't be surprised if this was an artifact of diffraction and that's what I am referring to.

Asher
 
I do think that most of the talk about "diffraction limits" is exaggerated, or misapplied, though.

I'd say that its effects are mostly misunderstood, which can lead to odd conclusions.

Higher pixel density never, ever makes diffraction worse!

While essentially true, it doesn't help those who misunderstand the issues at hand. Per pixel microdetail does suffer from higher sampling density, even though the diffraction pattern stays the same for a given aperture! I've posted an example of the effect here.

Still, for a given pixel size, one shouldn't ignore the aperture.

Indeed, unless one has enough oversampled pixels to still satisfy a certain output size, even after downsampling or when viewing from a larger distance, the per pixel microdetail suffers from using apertures that are too narrow.

Bart
 
Bart,

I come across this term "per pixel" detail or microdetail farily often. I am an imaging scientist and I have never seen a technical definition of this term. Is there an agreed scientific meaning, or is this a folk-term?

-Nat
 

John Sheehy

New member
John, You made me laugh! Yes that's true, one could have, after all one huge pixel and that may well degrade the image to oblivion.

Well, yes, but you don't have to go that far to have the effect. It is relevant at all (currently) practical pixel pitch ranges. Your hypothetical light bleeding over into the next pixel because its closer because of a smaller pixel pitch is not realistic. In order for that to be an issue, the photons would already have to be aligned in a grid, at a precise ratio to the pixel pitch. You have a better chance of winning Powerball 5x in a row with the same numbers each time. In the real world, photons are all over the place, and that photon that might have landed right in the center of a pixel with a small pitch may have landed at the far edge of a larger pixel which it would otherwise have missed, winding up way off course. On the average, more photons will be more displaced with bigger pixels.

Still, for a given pixel size, one shouldn't ignore the aperture. Look at the 100% cut out of the image of the flower here. The aperture was f32 for a ~6.8 micron pitch sensel. See how the green has been bled into the edge of the white petal. Well you might argue that the petal does indeed have a green edge, but I wouldn't be surprised if this was an artifact of diffraction and that's what I am referring to.

Regardless of what the artist believes, that image has a lot of sharpening. Converters lie. When they say "0" sharpening, "0" is an arbitrary amount of sharpening. The image also has a large amount of noise reduction (and edge-preserving sharpening was probably used there, as well).

Also, we don't have the same shot at f/11, and/or the same optics with a different pixel pitch, so the effects of diffraction can be pure speculation.

But, yes, if you know you have big pixels, you know that they will limit the sharpness to the point where the losses due to diffraction are relatively minor - the correct and practical way of looking at things. :)
 

John Sheehy

New member
Bart,

I come across this term "per pixel" detail or microdetail farily often. I am an imaging scientist and I have never seen a technical definition of this term. Is there an agreed scientific meaning, or is this a folk-term?

It simply refers to aspects of images when viewed at a fixed pixel magnification, and where the resolution of the pixel is well above the threshold of resolution.

There seems to be a love affair between many people and high pixel-to-pixel contrast; a sort of eye-candy, but the imaging is defective or fragile. Any capture that resolves a maximum contrast black to white transition in less than about 2.5 - 3 pixels exclusive will have artifacts, and the popular desired effect of high pixel level contrast is fragile, depending on luck of alignment, and it disappears with any resampling, whether it be image resizing, rotation, perspective correction, CA removal, etc. The low-frequency aliasing artifacts will still be intact, though.
 

Asher Kelman

OPF Owner/Editor-in-Chief
John,

You are correct that with the flower image color smearing, (with the Phase One back at f32), we need more f stops tested and that was requested.

While we are at image degradation, I'm puzzled just how the CNET tester found image to be degraded going from 10 to 14MP in the Sony Alpha line of cameras. The new Alpha A350 really got slammed. I'm disappointed that they provide so little data with their opinions.

Reported in OPF here.

Asher
 

John Sheehy

New member
John,

You are correct that with the flower image color smearing, (with the Phase One back at f32), we need more f stops tested and that was requested.

While we are at image degradation, I'm puzzled just how the CNET tester found image to be degraded going from 10 to 14MP in the Sony Alpha line of cameras. The new Alpha A350 really got slammed. I'm disappointed that they provide so little data with their opinions.

It could be that the comments are based on images with noise reduction. The fact that monitor pixel pitches run in such a small range makes it impossible to view high-MP images fairly on the monitor, or even in prints with low-PPI print drivers drives manufacturers to reduce noise (and consequently, detail) as if the only purpose of higher pixel densities is larger display sizes.
 
Bart,

I come across this term "per pixel" detail or microdetail farily often. I am an imaging scientist and I have never seen a technical definition of this term. Is there an agreed scientific meaning, or is this a folk-term?

-Nat

Hi Nat,

As stated in this context, it's a 'folk-term'. It attempts to distinguish between total 'image-resolution' and 'pixel-resolution', which both are important. Technically one needs at least 2 pixels to describe their contrast.

Image-resolution takes into account that if we reproduce the individual pixels small enough, use a high PPI setting on a printer, downsample the image, or look at it from a longer distance, the resolution may match or exceed our visual acuity at a given viewing distance.

Per pixel resolution refers to the potential to reproduce microdetail, contrast differences between adjacent pixels that allow to reproduce very fine surface structures and thin features e.g. twigs at a distance.

The pixel resolution is more scientificly described by Point Spread Functions (PSF) and derived Modulation Transfer Functions (MTF). PSFs and MTFs are not everybody's cup of tea, yet they are the only unambiguous way to describe component/system performance.

The MTFs describe the contrast reproduction as a function of feature size. It allows to predict how well a system can reproduce small contrast differences (microdetail) at the limiting resolution, the contrast difference between 2 pixels. If a system reduces contrast to 10% of its original contrast, then it follows that a 10% subject contrast gets reduced to 1% final contrast, which happens to be approx. the limiting human visual resolution. Thus we can predict how well minute contrast differences can be reproduced, and how life-like the image will look.

There are also other factors that influence realism in image reproduction, and some of them are also decribed by the MTF curve.

Bart
 

David Sommars

New member
I'm sure the digic III is a nice thing to have but the sensor is the same size as the old rebels, 12mp isnt going to make your pictures much nicer if any in this case.

save your money towards a nice lens and/or the 5D upwards....

If your just getting into it, this is a nice camera, but still, I see sensor size a huge limiting factor. dont be bedazzled by MP figures, its bloat
 

John Sheehy

New member
I'm sure the digic III is a nice thing to have but the sensor is the same size as the old rebels, 12mp isnt going to make your pictures much nicer if any in this case.

Well, the current XTi has fairly poor quantum efficiency, like the 5D, the older 1-series cameras, the 10D, etc, and will probably improve unless Canon is intentionally holding it back. Native ISO can increase from the ~ISO 87 that the 400D has, to about 120 like the more efficient, newer cameras (20D and up, and 1Dmk3) cameras. Read noise always has room for improvement, but even if the pixel level read noise does not improve in the XSi, 20% more pixels will translate into a 9% decrease in image-level read noise. More recent cameras also seem to have less line-noise (aka banding), which increases the usability of shadows.

It all depends on context, what the relative value of this camera will be. If you have your longest lens mounted on a tripod, and need to resolve a small subject as best as possible, the XTi will give the most resolution, and will have less noise than any of the FF or 1.3x Canon DSLRs, with the possible exception of the 1DSmk3.

If you could stick any of the better P&S camera bodies on the lens and the subject wasn't too big for the frame, they would do better yet.

No, I'm not crazy; I am absolutely correct.

save your money towards a nice lens and/or the 5D upwards....

5D is old tech. It surpasses the crop cameras when you fill your frame the same way, but is still a step behind the newer Canon Pro DSLRs, and the Nikon D3. The 5D replacement will undoubtably have better full-frame quality than the 5D, but it will still take second place to the XSi with a fixed lens choice, and subject that fits the XSi frame. It would take 31MP to equal the XSi in absolute resolution.

If your just getting into it, this is a nice camera, but still, I see sensor size a huge limiting factor. dont be bedazzled by MP figures, its bloat

Larger sensors can give lower image shot noise when you fill the frame the same way, but sensor resolution is real when you have the optics to back it. An XSi with one of the sharper lenses will get more true, unaliased subject detail than any FF with less pixels, regardless of the lens on the FF.
 

David Sommars

New member
I dont see how the sensor resolution can really be much better when the sensor is the same size and the only improvement is updated processor and more pixels are crammed into the same area.

Isnt this the same as adding resolution in photoshop ?
I could be wrong, just my gut feeling.

I mean if all you needed to do was update the processor and cram more pixels into the same size mm sensor why does the the hassies and sinar use sensors that are 2 or 3 times the size of the XSI ?

I think the limiting factor in most digital is the tonal range which I attribute to the bit depth, and more largely to the sensor size. who knows, maybe in wrong...
 

John Sheehy

New member
I dont see how the sensor resolution can really be much better when the sensor is the same size and the only improvement is updated processor and more pixels are crammed into the same area.

"Crammed" is a connotatively loaded word. The evidence suggests that "cramming" is good, in general. The best 10/12 MP P&S cameras "cram" a lot more pixels than the densest DSLRs, and yet collect far more photons per unit of area with the same exposure (1.5x - 3x as many). The crammed little sensors have less shot noise per unit of area than the DSLRs at all ISOs. They also have far less read noise per unit of area than all CCD DSLRs, and all CMOS DSLRs at low ISOs. On the cleaner DSLRs (like the 1D mk3 cameras and the D3), the crammed little sensors have about the same or slightly inferior read noise per unit of area at ISO 1600.

Isnt this the same as adding resolution in photoshop ?

Not at all; "adding resolution in Photoshop" does not make the noise grain or bayer artifacts finer, and does not decrease overall image read noise the way higher pixel density can. It can also never give more subject detail.

I could be wrong, just my gut feeling.

Your gut may not have a useful working model.

I mean if all you needed to do was update the processor and cram more pixels into the same size mm sensor why does the the hassies and sinar use sensors that are 2 or 3 times the size of the XSI ?

Digital backs are made to replace film on a format in which people already have a huge lens and body investment. They are *extremely* inefficient at collecting photons, in general. Most have no microlenses, and lose most of the photons between the photosites. They also tend not to have anti-alias filters as well, to give the "pixel-sharp" look I spoke negatively of earlier in the thread. Someone in another web forum I read linked to RAWs from the 22MP Sinar; the pixels only collected twice as many photons with the same exposure as the pixels in the 10MP Panasonic FZ50, and had higher read noise. There's 22M of them, and that is their only savior, besides the fine lenses they get to work with. The same space filled with pixels from the FZ50 would give vastlly superior results. The fact that the back offers ISO 25 is a hint; there is no way, with current technology, for a single exposure of the sensor to give ISO 25 with an efficient photon collection, unless that ISO 25 had no headroom (too little to work with).

I think the limiting factor in most digital is the tonal range which I attribute to the bit depth, and more largely to the sensor size. who knows, maybe in wrong...

Bit depth is generally not an issue at this point in time. The Canon K10D at ISO 100 is the only camera I am aware of that actually has even slightly posterized RAW data (and ever so slightly).

Sensor size, in theory, should place a limit on total-image-level S/N for photon shot noise. As I've already suggested, however, the efficiency varies greatly amongst cameras, and it is always possible for a smaller sensor to collect more photons through better efficiency.

The fact is, in the current state of affairs, the efficiency is highest for the smallest sensors, and lowest for the largest sensors, compressing their range of relative effectiveness much more than need be.
 

David Sommars

New member
If digibacks are designed to replace film then why do film cameras cost a few hundred bucks now and digi backs tens of thousands.

I dont know about photons or anything like that, but I just disagree on your whole idea based on feeling alone... lol !!!!!

Im sure the newer dslrs are very efficient, I'm thinking in terms of total tonality. You aren't suggesting the XSI has better dynamic range then the 5D are you ?


the guy comment is hilarious BTW !!
 

John Sheehy

New member
If digibacks are designed to replace film then why do film cameras cost a few hundred bucks now and digi backs tens of thousands.

Because they make a small number of them, and people are willing to pay. They are not wonders of technology, by any means.

I dont know about photons or anything like that, but I just disagree on your whole idea based on feeling alone... lol !!!!!

Im sure the newer dslrs are very efficient, I'm thinking in terms of total tonality. You aren't suggesting the XSI has better dynamic range then the 5D are you ?

The 400D has slightly higher DR than the 5D at ISO 100, even though the 5D has a little less noise in the highlights. It is not unlikely that the XSi improves in this regard at base ISO. It translates into highlight headroom more with the 400D because it meters 1/2 stop further from RAW clipping than the 5D.
 
Dear John,

May I intervene and correct a post which is partly right and partly also wrong/un-complete.

Sensors like the ones used by MFDB manufacturers have typically an efficiency between 70% and 80%: this means that 70% to 80% of the pixels of such CCD sensors are sensitive to light, and 20% to 30% are not. As such a sensor from the same built can be 20 to 30% more sensitive from another.

The Panasonic sensor is significantly smaller and has as well significantly smaller pixels: to get here somehow still some sensitivity it NEEDS to use micro-lenses. The use of micro-lenses, however, makes it impossible/difficult to use with a view camera with tilts/swings/shifts or with short focal length lenses.

Smaller pixels have in addition the negative effect/disadvantage that the dynamic range (DR) is much smaller, that those pixels are faster "full", thus leading very quickly to over-exposure. Also details in the shadows get lost

There is of course much more to say about the differences, and things are not that simple as often presented.

Best regards,
Thierry

Digital backs are made to replace film on a format in which people already have a huge lens and body investment. They are *extremely* inefficient at collecting photons, in general. Most have no microlenses, and lose most of the photons between the photosites. They also tend not to have anti-alias filters as well, to give the "pixel-sharp" look I spoke negatively of earlier in the thread. Someone in another web forum I read linked to RAWs from the 22MP Sinar; the pixels only collected twice as many photons with the same exposure as the pixels in the 10MP Panasonic FZ50, and had higher read noise. There's 22M of them, and that is their only savior, besides the fine lenses they get to work with. The same space filled with pixels from the FZ50 would give vastlly superior results. The fact that the back offers ISO 25 is a hint; there is no way, with current technology, for a single exposure of the sensor to give ISO 25 with an efficient photon collection, unless that ISO 25 had no headroom (too little to work with).

The fact is, in the current state of affairs, the efficiency is highest for the smallest sensors, and lowest for the largest sensors, compressing their range of relative effectiveness much more than need be.
 
Smaller pixels have in addition the negative effect/disadvantage that the dynamic range (DR) is much smaller, that those pixels are faster "full", thus leading very quickly to over-exposure. Also details in the shadows get lost

Hi Thierry,

I fully agree, and the loss of shadow detail is apparent, even in many correctly exposed small sensel images. Dynamic range requires sensel area to collect adequate charge above the noise floor. The 6.4 micron pitch sensels in my EOS-1Ds Mark III already require very accurate exposure (bracketing) to get the best DR out of the images, smaller sensels are plainly problematic.

Bart
 

Asher Kelman

OPF Owner/Editor-in-Chief
With large format cameras, all things otherwise equal, it is often asserted that shadow areas can lose detail with multicoated lenses.

If this is so, I wonder whether such a phenomenon could be as or even perhaps more important in small format digicams. Here the antireflective coatings on the modern camera lens and also on micro-lenses may be decreasing dynamic range further than expected. If this is true, then the MF systems, without micro lenses and also the Foveon CMOS chip, again with no micro-lenses might have less dynamic range degradation by virtue of escaping from layers of coatings.

Asher
 
With large format cameras, all things otherwise equal, it is often asserted that shadow areas can lose detail with multicoated lenses.

If this is so, I wonder whether such a phenomenon could be as or even perhaps more important in small format digicams.

Multi-coating increases transmission by reducing reflection. Internal reflections in the lens will reduce the dynamic range of the projected image. The internal reflections will be most noticable in the shadows as added luminance, but by reducing contrast! While that may somewhat brighten the shadows, it won't help the detail.

Bart
 
Hi Nat,

As stated in this context, it's a 'folk-term'. It attempts to distinguish between total 'image-resolution' and 'pixel-resolution', which both are important. Technically one needs at least 2 pixels to describe their contrast.

Image-resolution takes into account that if we reproduce the individual pixels small enough, use a high PPI setting on a printer, downsample the image, or look at it from a longer distance, the resolution may match or exceed our visual acuity at a given viewing distance.

Per pixel resolution refers to the potential to reproduce microdetail, contrast differences between adjacent pixels that allow to reproduce very fine surface structures and thin features e.g. twigs at a distance.

The pixel resolution is more scientificly described by Point Spread Functions (PSF) and derived Modulation Transfer Functions (MTF). PSFs and MTFs are not everybody's cup of tea, yet they are the only unambiguous way to describe component/system performance.

The MTFs describe the contrast reproduction as a function of feature size. It allows to predict how well a system can reproduce small contrast differences (microdetail) at the limiting resolution, the contrast difference between 2 pixels. If a system reduces contrast to 10% of its original contrast, then it follows that a 10% subject contrast gets reduced to 1% final contrast, which happens to be approx. the limiting human visual resolution. Thus we can predict how well minute contrast differences can be reproduced, and how life-like the image will look.

There are also other factors that influence realism in image reproduction, and some of them are also decribed by the MTF curve.

Bart
Hi Bart,

Sorry it has taken me so long to reply. Of course, I take your points about the psf and mtf. But I cringe a bit when people throw around the "folk-term" per pixel resolution, or similar, as though it had some well defined meaning. As you pointed out, the evaluation of resolution and contrast is a well worn path. The problem with the per pixel notion is that one is sometimes responding to what one sees without regard to a proper reference. For example, an image with much higher sampling, say a 1DsIII versus a 20D, may look less sharp at the pixel level than one with less megapixels while still having better resolution and contrast. I am trying here to differentiate between the digital sampling, viewing at 100% on a monitor, and actual resolution and contrast measures.

-Nat
 

John Sheehy

New member
Multi-coating increases transmission by reducing reflection. Internal reflections in the lens will reduce the dynamic range of the projected image. The internal reflections will be most noticable in the shadows as added luminance, but by reducing contrast! While that may somewhat brighten the shadows, it won't help the detail.

With digital it is genearlly a loss in IQ, because not only are the shadows low in contrast, but they now have more shot noise because there are more total photons collected, so when you apply the curve to make blacks black, the deep shadows have more noise than if they were optically close to black to begin with.

With film, I believe it can actually help to have a weak light blanket in the shadows, because the first few photons don't really register.
 
With film, I believe it can actually help to have a weak light blanket in the shadows, because the first few photons don't really register.

Correct, that's probably where Asher's sources got their assertion from. However, microdetail often has low contrast, and film also has low contrast near it's limiting resolution, so it might benefit some medium frequency shadow structures or high contrast microdetail more than subtle detail. So, as usual, it depends ...

Bart
 
Top