Native vs Downsampled?

Which image usually looks better, a native 5MP image from a 5MP camera or a downsampled 5MP imagine (M1) from an 8MP camera?
I think the downsampled looks worse.

1st image downsampled to 5MP (M1) on a Canon A630.



2nd image native (full) resolution 5MP from A610.



The latter certainly looks better. Both images were taken in Normal JPEG mode.

downsampling used with care has generally some advantages:
smaller file ;
greater sharpness;
a bit less noise ;
reduces artifacts.

but it has no advantages when resizing with a 8mpix image to 5 megapix because:
8mp 3264*2448 and 5mp 2592*1944.
so you resize about every 5*5 pizel to 4*4 pixels.

that results pretty unsharp images, it does not enhance picture quality in any way.

but you may get very pleasant results when resizing about every 3*3 pizels to 2*2.
the calculated result (2176*1498) is close to the A630's M2 setting
so use M2 resolution, that's pretty sharp

just for example for the a650 the M2 setting is the most optimal too, because of the (almost)same downsample procent. it has 5MP resolution, which is enough generally for non-resolution-critical purposes (prints below a3, wallpaper on a regular screen)
for later cropping, post processing use native resolution

(i own a630 and a650)
--
P0WERSH0T with: jpeg&RAW + live histogram and more...
Q: how? A: CHDK! Q: free? A: YES! Q: risks? A: NO!

How did you downsample? Did you use Photoshop bicubic sharper? That will help some (using bicubic sharper) but I have found it is still necessary to add sharpening to all downsampled images.

Downsampling softens, even if you're downsampling to the preferred sizes (50%, 33%, 25%, etc.).

Leroy

PineyPhotographer said:

The latter certainly looks better.

Click to expand...

I'm very interested in this question you're asking because I'm using my G10 at M1.

About your pictures... I don't see reasons to choose one or the other; they look similar to me.

Greetings

--
http://www.fotolog.com/pitquim

Due to my experience a downsized picture always looks better then a native picture of the same dimensions (assuming the larger picture is about the same quality when comparing by pixel peeping the larger picture with the native smaller picture). It is probably strongly depending on the resizing agorithm - in may case supersampling.

The downsampled picture is always sharper, with better resolution and with more details. Regardles the resizing ratio - i.e. for example also an 8Mpx picture resized to 5Mpx looks significantly better then a native 5 Mpx picture.

Anybody might disagree to any reason, byt this is may experience - I very often do resampling just for comparison purposes - the result is ALWAYS like this.

From my point of view this is also a solution for the nowadays cameras with high Mpx - just downsample the pictures to a size of e.g. 5Mpx and you get in most cases a better image quality then with older cameras with lower Mpx - in this case 5Mpx. Sometimes there are exceptions - just this morning I compared a downsampled Canon G10 picture (resized to 6Mpx) with a native 6Mpx picture from Canon G6 and the result was not very persuasive - the G6 picture simply looked better.

Sano

http://blog.dpreview.com/editorial/2008/11/downsampling-to.html

--
http://www.fotolog.com/pitquim

spare time:
http://forums.dpreview.com/forums/read.asp?forum=1000&message=30147854

While Phil's tests are interesting, his methodology and conclusions seem controversial, to say the least. cheers, gkl

pitquim said:

http://blog.dpreview.com/editorial/2008/11/downsampling-to.html

Click to expand...

gkl said:

While Phil's tests are interesting, his methodology and conclusions
seem controversial, to say the least. cheers, gkl

pitquim said:

http://blog.dpreview.com/editorial/2008/11/downsampling-to.html

Click to expand...

Click to expand...

That discussion gets pretty deep.

I did a test too. I took the ISO 800 G10 RAW image from Imaging Resource and ran some noise numbers at various reduced sizes.

My results did not match Phil's. In Phil's article the noise hit a plateau after downsizing to 10MP. I didn't find a plateau. Maybe I did something wrong.

I used Noise Ninja and Noiseware to get noise numbers and a noise graph for each image (see the legend below). First I downsized to 50% with PS bicubic sharper and the noise numbers increased which makes sense. All of the other resizes were done with PS bicubic (plain).

The noise numbers/graphs were the results for the whole image, not a crop of the image. There was no editing beyond resizing (no sharpening).

So here are the results for the full size image, 2 50% resizes and a resize to 900 pixels:



And here is a chart of that first Noise Ninja number at the various image widths (a 25% resize is included in this chart which is not pictured above):



Legend:



Leroy

lwiley said:

I did a test too. I took the ISO 800 G10 RAW image from Imaging
Resource and ran some noise numbers at various reduced sizes.

My results did not match Phil's. In Phil's article the noise hit a
plateau after downsizing to 10MP. I didn't find a plateau. Maybe I
did something wrong.

Click to expand...

lwiley said:

Click to expand...

lwiley said:

I used Noise Ninja and Noiseware to get noise numbers and a noise
graph for each image (see the legend below).

Click to expand...

Thanks for posting your results. Comparing to:
http://blog.dpreview.com/editorial/2008/11/downsampling-to.html
it looks like you went to much smaller sizes than Phil. His graph stops at 50%.

It looks like you both measure similar noise reduction (about 20%) for PS Bicubic over his narrow range. So, if I am reading your graphs right, it looks like your results are consistent with his.

Of course, the real issue with Phil's blog is the following statement:

"At the end of it all downsampling is no substitute for larger sensors or larger photosites."

While it is generally true that larger sensors give better noise performance for a given field of view, the dependence of noise on photosites is more complex than just size alone.

Phil's noise blog was apparently intended to prove that you don't get much reduction in noise by downsampling the 14.6 MP G10 to 10 MP LX3 image size. But by the same token, DPR's graphs show that the G10 is not much noisier than the LX3 to begin with. The results of Phil's noise blog just reinforce that there is very little difference between the high ISO performance of the G10 and LX3. However, there is a significant difference in the low ISO performance. cheers, gkl

gkl said:

lwiley said:

I did a test too. I took the ISO 800 G10 RAW image from Imaging
Resource and ran some noise numbers at various reduced sizes.

My results did not match Phil's. In Phil's article the noise hit a
plateau after downsizing to 10MP. I didn't find a plateau. Maybe I
did something wrong.

Click to expand...

Click to expand...

gkl said:

gkl said:

I used Noise Ninja and Noiseware to get noise numbers and a noise
graph for each image (see the legend below).

Click to expand...

Thanks for posting your results. Comparing to:
http://blog.dpreview.com/editorial/2008/11/downsampling-to.html
it looks like you went to much smaller sizes than Phil. His graph
stops at 50%.
It looks like you both measure similar noise reduction (about 20%)
for PS Bicubic over his narrow range. So, if I am reading your graphs
right, it looks like your results are consistent with his.

Click to expand...

You're right, the results are similar. I did not look close enough at the graph. Also, the PS Bicubic is the one line on the graph that does not show a plateau.

gkl said:

Of course, the real issue with Phil's blog is the following statement:
"At the end of it all downsampling is no substitute for larger
sensors or larger photosites."
While it is generally true that larger sensors give better noise
performance for a given field of view, the dependence of noise on
photosites is more complex than just size alone.
Phil's noise blog was apparently intended to prove that you don't get
much reduction in noise by downsampling the 14.6 MP G10 to 10 MP LX3
image size. But by the same token, DPR's graphs show that the G10 is
not much noisier than the LX3 to begin with. The results of Phil's
noise blog just reinforce that there is very little difference
between the high ISO performance of the G10 and LX3. However, there
is a significant difference in the low ISO performance. cheers, gkl

Click to expand...

Agree here too. There is not that much difference in the high ISO noise. I will be living in the low ISO world 95% of the time, where the G10 does very well.

Leroy