Today's lesson: Why up-sampling lo-res data to compare it to high-res data is bogus.

Problem: online sites often compare lo-res digital images to high-res digital images using a common technique: up-sampling the lo-res data (typically in Photoshop using bi-cubic sampling) to match the high-res data so that it can be presented "objectively" on the web.

This technique is presented as being absolutely objective and promoted as a fair comparison. And to date, few have challenged the methodology. Herein I attempt to present the reasons why the comparison is both false and misleading (the effect of the technique is to always make the lower-res data look significantly better than it really is, regardless of the type of photo)

Let's take a simple piece of data, such as the following letter "T":

Figure A: high-res data "as is"

The above has been scaled WITHOUT SAMPLING to 200% of original size so that you can see with clarity the detail in the pixels.

Now let's do the same for the same data only at precisely 1/2 the resolution:

Figure B: lo-res data "as is"

The comparison is immediate and dramatic. It is CLEAR to the human eye that one image has twice the resolution of the other. If online knowledge clearing houses such as Luminious Landscape would use such comparisons, they would inevitably draw different conclusions about the "quality" of data being delivered by the equipment they were reviewing. (OTOH, by doing so, they might end up upsetting their advertisers and sponsors. So it may be no coincidence that this has become the de facto "proof" offered to describe the relative merits of modern digital cameras vis a vis other -- ostensibly better -- technologies)

But of course, they don't present the data as I have shown here. Instead they upsample the low-res data so that it matches the resolution of the better data. The effect is equally dramatic (but misleading):

Figure C: lo-res up-sampled

While MR et al did not perform the next step in their much balleyhooed online showdown of film vs. digital (at least they said they didn't), many many other sites then take the above "C" data and sharpen it as such:

Figure D: lo-res up-sampled, sharpened

Consequently, most sites end up comparing figure A with figure D and then go on to conclude: "while A was sharper than D, it was only slightly so". This in spite of the fact that in reality A has double the resolution!!! Their inevitable conclusion? D is "almost" as sharp as A, conveniently neglecting to reveal figure B to their readers.

You, of course, get to draw your own conclusions. But as consumers, we should demand more from the sources we rely upon for infomation.

OTHER ONLINE SOURCES RELATED TO THIS ARTICLE

BetterLight: The Rest of the Picture

Clarkvision: Film vs. Digital Summary

J MICHAEL SULLIVAN Photography
	Today's lesson: Why up-sampling lo-res data to compare it to high-res data is bogus. Problem: online sites often compare lo-res digital images to high-res digital images using a common technique: up-sampling the lo-res data (typically in Photoshop using bi-cubic sampling) to match the high-res data so that it can be presented "objectively" on the web. This technique is presented as being absolutely objective and promoted as a fair comparison. And to date, few have challenged the methodology. Herein I attempt to present the reasons why the comparison is both false and misleading (the effect of the technique is to always make the lower-res data look significantly better than it really is, regardless of the type of photo) Let's take a simple piece of data, such as the following letter "T": Figure A: high-res data "as is" The above has been scaled WITHOUT SAMPLING to 200% of original size so that you can see with clarity the detail in the pixels. Now let's do the same for the same data only at precisely 1/2 the resolution: Figure B: lo-res data "as is" The comparison is immediate and dramatic. It is CLEAR to the human eye that one image has twice the resolution of the other. If online knowledge clearing houses such as Luminious Landscape would use such comparisons, they would inevitably draw different conclusions about the "quality" of data being delivered by the equipment they were reviewing. (OTOH, by doing so, they might end up upsetting their advertisers and sponsors. So it may be no coincidence that this has become the de facto "proof" offered to describe the relative merits of modern digital cameras vis a vis other -- ostensibly better -- technologies) But of course, they don't present the data as I have shown here. Instead they upsample the low-res data so that it matches the resolution of the better data. The effect is equally dramatic (but misleading): Figure C: lo-res up-sampled While MR et al did not perform the next step in their much balleyhooed online showdown of film vs. digital (at least they said they didn't), many many other sites then take the above "C" data and sharpen it as such: Figure D: lo-res up-sampled, sharpened Consequently, most sites end up comparing figure A with figure D and then go on to conclude: "while A was sharper than D, it was only slightly so". This in spite of the fact that in reality A has double the resolution!!! Their inevitable conclusion? D is "almost" as sharp as A, conveniently neglecting to reveal figure B to their readers. You, of course, get to draw your own conclusions. But as consumers, we should demand more from the sources we rely upon for infomation. OTHER ONLINE SOURCES RELATED TO THIS ARTICLE BetterLight: The Rest of the Picture Clarkvision: Film vs. Digital Summary © 2006 J Michael Sullivan, all rights reserved. www.jmichaelsullivan.com