best downsizing for superior IQ

[wmchauncey]

I'm still pondering a Canon 5DsR, the 50 MP beast, in search of the ultimate IQ from the camera.

It is communally accepted that downsizing a high MP image to one of a lower one will produce a
better IQ than one would achieve by using a lower MP camera.

Question is, are you better off doing it in-camera as the Canon has the ability to drop down to a
1.3 or 1.6 crop in camera or, accomplish the whole thing in Photoshop by increasing the ppi?

BTW, what happens to noise levels with either technique?

[scyth]

crop...ppi

and what "crop" or "ppi" have to do with downsizing ?

[ErikKaffehr]

Hi,

Changing PPI is nonsense, sorry!

As a matter of fact, correct downsizing is not easy. The problem is that downsizing always induces aliasing. The best way to avoid aliasing is probably to pre blur the image before downsizing.

My understanding is that downsizing in LR works pretty well why Photoshop's built in function may be less good. But, the aliases from downsizing are in most cases no big problem.

Regarding noise, downscaling to half size means a quarter of the area. So, if you downsize a 50 MP image by half the resulting image will be 12.5 MP. SNR ratio will double.

Regarding SNR (Signal Noise Ratio), it is really a "zero sum game". A 24x36 mm surface exposed at f/8 and 1/125s will collect the same amount of photons weather the pixels are 9 microns across or 4.5 microns across, assuming fill factor is the same. Larger sensors may have a better photodiode area to pixel are ratio, but modern sensor have a pretty decent sensel area to writing ration and microlenses focusing the light on the photodiode also help.

Smaller pixels reduce aliasing, that is fake detail. So reducing pixel size you get less aliasing, that is more accurate sampling at the price of a small reduction of DR (Dynamic Range). For each kind of pixel design there will be an optimum pixel size which offers best combination of DR vs resolution and fake detail.

Advanced printers can use 720 or 600 PPI while 180 PPI may be enough for an excellent print, when viewed at a reasonable distance. So, you don't need to downscale for prints, really.

Just as an example, my Sony A7r has 7952 x 5304 pixels, I can print this image at 11" x 7.4" on my Epson SP 3880 at 720 PPI without scaling down the image. Or, I can make a 44"x29" at 180 PPI without doing an upprez. Just to say, you shouldn't view that 44"x29" closer than 20" for best viewing experiment.

Best regards
Erik

I'm still pondering a Canon 5DsR, the 50 MP beast, in search of the ultimate IQ from the camera.

It is communally accepted that downsizing a high MP image to one of a lower one will produce a
better IQ than one would achieve by using a lower MP camera.

Question is, are you better off doing it in-camera as the Canon has the ability to drop down to a
1.3 or 1.6 crop in camera or, accomplish the whole thing in Photoshop by increasing the ppi?

BTW, what happens to noise levels with either technique?

[Tim Lookingbill]

Erik, can you address the quality differences (aliasing and noise) the OP asks about between downsizing within the camera over doing it in Photoshop/ACR?

If doing it incamera shooting Raw I wonder if the camera's downsizing may be better because of the manufacturer's proprietary knowledge of its sensor design, circuitry and A/D converter.

Has anyone done an A/B test on this. I think I remember years back but with improvements to camera design and performance things might have changed to show an improvement in this area of downsizing pixels.

[TonyW]

I'm still pondering a Canon 5DsR, the 50 MP beast, in search of the ultimate IQ from the camera.

It is communally accepted that downsizing a high MP image to one of a lower one will produce a
better IQ than one would achieve by using a lower MP camera.

Question is, are you better off doing it in-camera as the Canon has the ability to drop down to a
1.3 or 1.6 crop in camera or, accomplish the whole thing in Photoshop by increasing the ppi?

BTW, what happens to noise levels with either technique?

If you are looking for the ultimate IQ would it be fair to say that this is because you intend to print at the highest quality such as sending highest ppi to printer ?  If not and you are only looking to view on a monitor (even a high resolution one) then I wonder if a 50 MP camera is of real benefit.

AFAIK (I do not shoot Canon) the Canon does not actually downsize at all it merely crops the image to 1.3 or 1.6 crop factor.  In other words it just discards the pixels outside of the 1.3 /1.6 (30 MP / 19 MP vs the FF 50 MP) crop leaving you with a smaller file size but no reduction in noise or increase in resolution in other words similar to an APS C camera. 

I think it also has M raw and S raw but I do not know much about these other than a guess that they may not be raw files in the truest sense but are again a reduced pixel count from the native resolution for when you do not require the max IQ benefits of the full 50 MP

Erik

Just as an example, my Sony A7r has 7952 x 5304 pixels, I can print this image at 11" x 7.4" on my Epson SP 3880 at 720 PPI without scaling down the image. Or, I can make a 44"x29" at 180 PPI without doing an upprez. Just to say, you shouldn't view that 44"x29" closer than 20" for best viewing experiment.

I know that 180 PPI for Epson or probably 150 PPI Canon/HP should give good results particularly at the viewing distance quoted but I would generally upprez to 300 PPI (Canon) printer as I have seen that it can give better detailed results certainly when you expect photographers to look at the print . 

I have found that even for the cheapest A4 Canon printers this applies still applies although the testing I did recently was comparing sending 100 ppi from a cropped image (native file resolution at A4) direct from a cropped image in LR vs choosing 300 PPI in LR - the difference quite marked.  I suppose its horses for courses but would you uprezz your 180 PPI to 360 PPI for that special print?

[Peano]

or, accomplish the whole thing in Photoshop by increasing the ppi?

Say what??? Increasing ppi doesn't downsize the image. It makes the image larger. Forget ppi. That isn't relevant to reducing the pixel dimensions of an image.

[Jack Hogan]

If doing it incamera shooting Raw I wonder if the camera's downsizing may be better because of the manufacturer's proprietary knowledge of its sensor design, circuitry and A/D converter.

Tim,

for CaNikon downsizing is done in software either way,  proprietary knowledge of the hardware makes virtually no difference in most practical situations.  What does make a big difference are the chosen algorithm and relative parameters, of either of which there are many.  The better ones are slowly evolving and processing power hungry, which makes me think that in-computer downsizing (as opposed to in-camera) will produce the better marginal results, just as in rendering.

Has anyone done an A/B test on this. I think I remember years back but with improvements to camera design and performance things might have changed to show an improvement in this area of downsizing pixels.

Take a look at this for a crude quantitative comparison.  The amount of perceived aliasing is related to the area under the curves above the Nyquist frequency.

Jack

[Jack Hogan]

Question is, are you better off doing it in-camera as the Canon has the ability to ...

Hi wm,

If you think that you'll be downsizing often consider getting the 5DS vs the 5DSR because the antialiasing filter in the 5DS is quite efficient and will give you a good start towards the final dowrezed photograph.

Jack

[ErikKaffehr]

Hi,

I am belonging to the school of thought who says that you should always send the image to the printer in the printers native resolution. That would be 360 or 720 PPI for Epson and 300/600 PPI for Canon.

The reason for that is that programs like Photoshop or Lightroom use more resource intensive interpolation routines than the printer driver. This may mostly effect rendition of edges.

I real world, I don't think that matters a lot. You need to print large and pixel peep to see the difference.

My take is: Use a recipe that has been declared good by the experts, like Jeff Schewe. It may be better to focus our efforts on other issues, like making a decent image to print…

Best regards
Erik

If you are looking for the ultimate IQ would it be fair to say that this is because you intend to print at the highest quality such as sending highest ppi to printer ?  If not and you are only looking to view on a monitor (even a high resolution one) then I wonder if a 50 MP camera is of real benefit.

AFAIK (I do not shoot Canon) the Canon does not actually downsize at all it merely crops the image to 1.3 or 1.6 crop factor.  In other words it just discards the pixels outside of the 1.3 /1.6 (30 MP / 19 MP vs the FF 50 MP) crop leaving you with a smaller file size but no reduction in noise or increase in resolution in other words similar to an APS C camera. 

I think it also has M raw and S raw but I do not know much about these other than a guess that they may not be raw files in the truest sense but are again a reduced pixel count from the native resolution for when you do not require the max IQ benefits of the full 50 MP

ErikI know that 180 PPI for Epson or probably 150 PPI Canon/HP should give good results particularly at the viewing distance quoted but I would generally upprez to 300 PPI (Canon) printer as I have seen that it can give better detailed results certainly when you expect photographers to look at the print . 

I have found that even for the cheapest A4 Canon printers this applies still applies although the testing I did recently was comparing sending 100 ppi from a cropped image (native file resolution at A4) direct from a cropped image in LR vs choosing 300 PPI in LR - the difference quite marked.  I suppose its horses for courses but would you uprezz your 180 PPI to 360 PPI for that special print?

[TonyW]

Hi,

I am belonging to the school of thought who says that you should always send the image to the printer in the printers native resolution. That would be 360 or 720 PPI for Epson and 300/600 PPI for Canon.

The reason for that is that programs like Photoshop or Lightroom use more resource intensive interpolation routines than the printer driver. This may mostly effect rendition of edges.

I real world, I don't think that matters a lot. You need to print large and pixel peep to see the difference.

My take is: Use a recipe that has been declared good by the experts, like Jeff Schewe. It may be better to focus our efforts on other issues, like making a decent image to print…

Best regards
Erik

Thanks Erik for your thoughts, its just that I was a little surprised during a recent quick play to see how much improvement made to perceived resolution when sending the image direct from LR at the files native resolution vs sending at the printer native resolution.  The improvement was such that I doubted my testing (TBH still do!) but the only thing that I could think of that would have changed is the sharpening applied by LR at the different PPI i.e. the radius would be increased for the higher PPI and maybe the amount ? 

The bottom line is that for this test when viewed at a 'reasonable' distance the comparison showed no resolution benefit that I could see between 100PPI (file native) or 300PPI (entry level Canon printer), but was very striking once viewed at 24 or closer". 

I do take very serious note of what people like Jeff Schewe say about these matters and tend to follow his example.

As to focus on other issues such as making a decent or worthwhile image to print, well that is probably the most difficult to achieve 

[JRSmit]

I'm still pondering a Canon 5DsR, the 50 MP beast, in search of the ultimate IQ from the camera.

It is communally accepted that downsizing a high MP image to one of a lower one will produce a
better IQ than one would achieve by using a lower MP camera.

Question is, are you better off doing it in-camera as the Canon has the ability to drop down to a
1.3 or 1.6 crop in camera or, accomplish the whole thing in Photoshop by increasing the ppi?

BTW, what happens to noise levels with either technique?

But will it be better than the originaliteit 50mp image quality?
If it is for printing the more good pixels the better. An epson printer is best when feeding it with 720 pixels per inch on paper. And yes you can see the difference compared to 360ppi print resulution.
Perhas focus should be on getting the best possible image quality out of the 50mp camera.

[ErikKaffehr]

Hi Tony,

I don't really understand your question.

Jeff suggests that 360 PPI or 720 PPI are used when printing on Epson. Epson has a high definition mode at 2880 DPI, if I recall correctly and that is needed for 720 PPI.

What Jeff suggested that 360 PPI should be used if native resolution is below 360 PPI and 720 PPI if native resolution is above 360 PPI.  With that processing Lightroom choses an output sharpening that was developed in cooperation with Jeff and the late Bruce Fraser.

360 PPI corresponds to resolution of 20/20 human vision at 10" (25 cm). 180 PPI is often mentioned to be sufficient for excellent prints. This corresponds to 20/20 vision at 20"/ 50 cm.

I made an experiment recently with a high performing lens and a less high performing lens on print size 31"x47". That corresponded pretty exactly to 180 PPI. Looking close, the difference between the two lenses was quite obvious but moving back to say 100 cm from the prints all the difference was gone. The 180 PPI print viewed at close (40-50 cm) looked pretty much like the screen at actual pixels.

Best regards
Erik

Thanks Erik for your thoughts, its just that I was a little surprised during a recent quick play to see how much improvement made to perceived resolution when sending the image direct from LR at the files native resolution vs sending at the printer native resolution.  The improvement was such that I doubted my testing (TBH still do!) but the only thing that I could think of that would have changed is the sharpening applied by LR at the different PPI i.e. the radius would be increased for the higher PPI and maybe the amount ? 

The bottom line is that for this test when viewed at a 'reasonable' distance the comparison showed no resolution benefit that I could see between 100PPI (file native) or 300PPI (entry level Canon printer), but was very striking once viewed at 24 or closer". 

I do take very serious note of what people like Jeff Schewe say about these matters and tend to follow his example.

As to focus on other issues such as making a decent or worthwhile image to print, well that is probably the most difficult to achieve 

[TonyW]

Hi Tony,

I don't really understand your question.

Jeff suggests that 360 PPI or 720 PPI are used when printing on Epson. Epson has a high definition mode at 2880 DPI, if I recall correctly and that is needed for 720 PPI.

What Jeff suggested that 360 PPI should be used if native resolution is below 360 PPI and 720 PPI if native resolution is above 360 PPI.  With that processing Lightroom choses an output sharpening that was developed in cooperation with Jeff and the late Bruce Fraser.

360 PPI corresponds to resolution of 20/20 human vision at 10" (25 cm). 180 PPI is often mentioned to be sufficient for excellent prints. This corresponds to 20/20 vision at 20"/ 50 cm.

I made an experiment recently with a high performing lens and a less high performing lens on print size 31"x47". That corresponded pretty exactly to 180 PPI. Looking close, the difference between the two lenses was quite obvious but moving back to say 100 cm from the prints all the difference was gone. The 180 PPI print viewed at close (40-50 cm) looked pretty much like the screen at actual pixels.

Best regards
Erik

Hi Erik,
Sorry I have not made it clear enough.  I understand calculating for 20/20 human vision and I use the formula PPI=3438/viewing distance and as I normally use either HP or Canon then my printer working figures are 300/600 PPI using 300 if native res. below and 600 if above.  I was however surprised when I made a couple of test prints recently.

Even though the OP has not been back I think it would be better if I start a new thread as this is really hijacking the original, and I will post some samples to make it clearer what I am trying to understand 

Just to update have now post new thread HERE. 
To the OP my apologies for latterly hijacking your thread - I just got a little carried away