Hot!!! Should the F707 be used at 2048x1536?

An Internet review of the F707 claimed the resolution had an efficiency of 75% at 5 MP, the highest seen for a consumer digital camera. I got to thinking about this and realized that any image size above the 75% was "empty magnification" and if the image is resampled by 75% no detail should be lost. One of the optional resolutions of the F707 is 2048x1536, which is 80% of the highest 2560x1920 option (the 75% & 80% are linear not area values). So, I did a test. I took a picture of the same width of newspaper print using a tripod and the built-in flash at 2560x1920 and 2048x1536, and for comparison a Kodak DC4800 (3 MP camera) at 2160x1440 (the DC4800 has a 3/2 ratio image). I then resampled the 2048x1536 and 2160x1440 images to 2560 pixels wide. The results below are cropped but not reduced portions of the images show there's not much difference between the F707 samples (the F707 in 2048x1536 option takes a 2560x1920 image and resamples in RAW mode 80%, so it's not like your using a 3 MP camera as shown by the DC4800 result). I'll have look into this further because it could result in a lot more images on the MemorySticks.

http://genji.image.pbase.com/u9/bobbobet/upload/1670774.ResolutionTest.jpg--Bobbo

If I understand what you're saying, you think you can take images at the lower res, then bump them up with an image editor and they appear the same? I disagree. I think the one you resampled has a lot more noise, or at least more noticeable noise, than the text has a greenish bloom on the edges.

Just my .02 though...

Bobbo said:

An Internet review of the F707 claimed the resolution had an
efficiency of 75% at 5 MP, the highest seen for a consumer digital
camera. I got to thinking about this and realized that any image
size above the 75% was "empty magnification" and if the image is
resampled by 75% no detail should be lost. One of the optional
resolutions of the F707 is 2048x1536, which is 80% of the highest
2560x1920 option (the 75% & 80% are linear not area values). So, I
did a test. I took a picture of the same width of newspaper print
using a tripod and the built-in flash at 2560x1920 and 2048x1536,
and for comparison a Kodak DC4800 (3 MP camera) at 2160x1440 (the
DC4800 has a 3/2 ratio image). I then resampled the 2048x1536 and
2160x1440 images to 2560 pixels wide. The results below are cropped
but not reduced portions of the images show there's not much
difference between the F707 samples (the F707 in 2048x1536 option
takes a 2560x1920 image and resamples in RAW mode 80%, so it's not
like your using a 3 MP camera as shown by the DC4800 result). I'll
have look into this further because it could result in a lot more
images on the MemorySticks.



--
Bobbo

Click to expand...

--Eric http://www.pbase.com/elamonthttp://www.thedigitalgalleries.com/ (a work in progress)

Bobbo,

I understand your point. Still, the true 5MP picture looks more detailed to my eyes. Perhaps this is because of quality of Sony downsampling algorithm, or jpeg compression of 3MP picture eats some details.
In any case, the true 5MP picture looks better to me.

--Vladimir.

I have to agree with ealomont and Vladimir. To my eye, the difference between the true 5mp shot and the image up-sampled from 2048x1536 is significant and easy to see. For a 5mp image, I would shoot at 2560x1920.

Howdy

Howdy said:

I have to agree with ealomont and Vladimir. To my eye, the
difference between the true 5mp shot and the image up-sampled from
2048x1536 is significant and easy to see. For a 5mp image, I would
shoot at 2560x1920.

Howdy

Click to expand...

I also agree that upsampling is a mistake if you have the chance to get full resolution right from the camera. Taking pixel from the image is qiete easy, but adding pixel is a case of guess something. No matter how good the software performs for that.

D.Jenett

You should photograph one of those standard resolution charts like Phil Askey uses in the reviews on this website.

I can't really tell the difference based on what you showed me. But, if the smaller JPEG takes up less space on the memory stick, then common sense dictates that SOMETHING is missing.

As we all know, the the F707 spits out a 5MP image where each pixel has data for three different color channels. But the reality is that each sensor on the CCD senses only one color, so the 5MP image really involves interpolation.

I suspect the difference between the lower resolution and the higher resolution isn't as great as most people would expect.

But if you really want to save space on the memory stick, why not just save in the standard mode instead of the fine mode? Double your jpegs that way.

I always save fine jpegs. Of course, with my Sony F707 in for repair for DLS, I'm not saving anything right now.

I suggest you try something with finer detail, as Bob D. has suggested. The larger 707 shot looks slightly sharper to me, but it's hard to tell with the absence of fine detail in the original image.

--Ron ParrFAQ: http://www.cs.duke.edu/~parr/photography/faq.htmlGallery: http://www.pbase.com/parr/

Let me clarify my post. I didn't say which image looks best, just that there wasn't much difference between the two F707 images. There are differences, some favor the 2560x1920 image and some favor the 2160x1440 image (look at the gap between the letters E & T in the word MARKET). As I stated, I have to look into this further.

The possibility, and I emphasize possibility, is that there is empty magnification in the 2560x1920 option because of the 75% resolution efficiency statement. If this is so, then the image may be resampled down to some degree and not loose any detail. If this is so, then using the 2048x1536 option may accomplish this. If this is so, I would not advocate taking pictures at 2048x1536 and resampling to 2560x1920, leave them at 2048x1536 because no detail would be added by enlargement. I resampled the 2048x1536 example for comparison with the 2560x1920 example because it is difficult to compare images that are different sizes. There will be some differences because the images are processed differently But,if the differences are very subtle this could be of major consequence.

Using the 2048x1536 option on the F707 is not taking a picture at 2048x1536, it's taking a 2560x1920 picture and resampling in the camera RAW mode to 2048x1536. A big difference, look at the DC4800 image.

I was surprised by the 75% resolution efficiency claim because I thought the highest value possible on a digital imager was 66.7%. I just never thought of the empty magnification consequence this presents.

"Empty magnification" is enlarging an image beyond the resolution limit. The image is larger but detail is not added.

Some of the "noise" is actually paper texture because it appears in the same place in the images.

Notice I use the words "possibility" and "if this is so". As I noted, I have to look into this further.
--Bobbo

Bobbo said:

An Internet review of the F707 claimed the resolution had an
efficiency of 75% at 5 MP, the highest seen for a consumer digital
camera.

Click to expand...

I did a similar test using Photoshop to overlay the blown-up 2048 over a full size 2560 of the same thing.

Clicking the blow up layer on and off, the actual visible difference was so small that nobody could see it except in a certain very few pixels. And then, the difference was very, very minor.

I shoot in 2048 most of the time. Because the files are smaller and the loss of detail, quality and print-ability is virtually nil.

And besides... when you open up a 2048 image on your computer screen, it is already pixel-perfect and looks better.

-iNova

I did this experiment too a couple of months ago.

Shot in bright daylight, two images - one in 3.3 MP the other in 5 MP both with the 707 camera. The differences seen in big enlargements are really small. Some extra background noise and a tiny loss of resolution.

When I compared the same pictures to similar ones shot with my Nikon CP990 I could see a much bigger difference.

Can't figure out why, but maybe I'm missing something somewhere. I do not dare to shoot at 3.3 MP from my 707 being afraid that I've overlooked something.

I chickened out and purchased 4 more 128 MB MS cards (have now 8 altogether) and shoot at 5 MP.

Yossi

Peter iNova said:

Bobbo said:

An Internet review of the F707 claimed the resolution had an
efficiency of 75% at 5 MP, the highest seen for a consumer digital
camera.

Click to expand...

I did a similar test using Photoshop to overlay the blown-up 2048
over a full size 2560 of the same thing.

Clicking the blow up layer on and off, the actual visible
difference was so small that nobody could see it except in a
certain very few pixels. And then, the difference was very, very
minor.

I shoot in 2048 most of the time. Because the files are smaller and
the loss of detail, quality and print-ability is virtually nil.

And besides... when you open up a 2048 image on your computer
screen, it is already pixel-perfect and looks better.

-iNova

Click to expand...

Yossi said:

I did this experiment too a couple of months ago.
Shot in bright daylight, two images - one in 3.3 MP the other in 5
MP both with the 707 camera. The differences seen in big
enlargements are really small. Some extra background noise and a
tiny loss of resolution.
When I compared the same pictures to similar ones shot with my
Nikon CP990 I could see a much bigger difference.
Can't figure out why, but maybe I'm missing something somewhere.

Click to expand...

This is very interesting but I don't think you are missing anything. It's a result of the color and luminance interpolation inherint in traditional color CCDs. This wouldn't be the case with Foveon's new chip.

Yossi Kramer said:

I do not dare to shoot at 3.3 MP from my 707 being afraid that I've
overlooked something.
I chickened out and purchased 4 more 128 MB MS cards (have now 8
altogether) and shoot at 5 MP.

Yossi

Peter iNova said:

Bobbo said:

An Internet review of the F707 claimed the resolution had an
efficiency of 75% at 5 MP, the highest seen for a consumer digital
camera.

Click to expand...

I did a similar test using Photoshop to overlay the blown-up 2048
over a full size 2560 of the same thing.

Clicking the blow up layer on and off, the actual visible
difference was so small that nobody could see it except in a
certain very few pixels. And then, the difference was very, very
minor.

I shoot in 2048 most of the time. Because the files are smaller and
the loss of detail, quality and print-ability is virtually nil.

And besides... when you open up a 2048 image on your computer
screen, it is already pixel-perfect and looks better.

-iNova

Click to expand...

Click to expand...

Yossi said:

When I compared the same pictures to similar ones shot with my
Nikon CP990 I could see a much bigger difference.
Can't figure out why, but maybe I'm missing something somewhere.

Click to expand...

What you are missing is the fact that the CP990 gives you 3 million pixels output, with each pixel having a value for red, green, and blue.

But actually, the CP990 doens't have 9 million sensors, it only has 3 million sensors, some of them green, some of them red, and some of them blue.

Thus the 3 million pixel image that comes from the CP990 is interpolated up. The internal algorithms are guessing what the R,G&B values should be for each pixel based upon the values from neighboring pixels.

And THAT is why the 3MP image from the 5MP F707 is sharper and has more detail than the 3MP image from the 3MP CP990.

Also, the F707 has a better lens. But I think that most of the difference in sharpness is due to pixel interpolation and not the lens.

Yossi said:

I did this experiment too a couple of months ago.
Shot in bright daylight, two images - one in 3.3 MP the other in 5
MP both with the 707 camera. The differences seen in big
enlargements are really small. Some extra background noise and a
tiny loss of resolution.
When I compared the same pictures to similar ones shot with my
Nikon CP990 I could see a much bigger difference.
Can't figure out why, but maybe I'm missing something somewhere. I
do not dare to shoot at 3.3 MP from my 707 being afraid that I've
overlooked something.

Click to expand...

If you read my threads in this message you'll see why this is so. The resolution efficiency of the F707 is about 75% at 5MP. If you compare your CP990 images taken at 2MP and 3MP you should get the same results, little difference between them. This effect is probably true for all digital cameras.

Yossi Kramer said:

I chickened out and purchased 4 more 128 MB MS cards (have now 8
altogether) and shoot at 5 MP.

Click to expand...

Memory card makers and retailers would probably not like this information known to users.
--Bobbo

from a practical standpoint ,without upsampeling does this realy make a difference if we are just talking about minor croping and printing at 8x10 size?

Before deciding to change out image size to 3mp equivalent, don't we need to examine some color samples and not just B/W text?

Bobbo said:

An Internet review of the F707 claimed the resolution had an
efficiency of 75% at 5 MP, the highest seen for a consumer digital
camera. I got to thinking about this and realized that any image
size above the 75% was "empty magnification" and if the image is
resampled by 75% no detail should be lost. One of the optional
resolutions of the F707 is 2048x1536, which is 80% of the highest
2560x1920 option (the 75% & 80% are linear not area values). So, I
did a test. I took a picture of the same width of newspaper print
using a tripod and the built-in flash at 2560x1920 and 2048x1536,
and for comparison a Kodak DC4800 (3 MP camera) at 2160x1440 (the
DC4800 has a 3/2 ratio image). I then resampled the 2048x1536 and
2160x1440 images to 2560 pixels wide. The results below are cropped
but not reduced portions of the images show there's not much
difference between the F707 samples (the F707 in 2048x1536 option
takes a 2560x1920 image and resamples in RAW mode 80%, so it's not
like your using a 3 MP camera as shown by the DC4800 result). I'll
have look into this further because it could result in a lot more
images on the MemorySticks.



--
Bobbo

Click to expand...

Hi,

I always shot at 3.3 mp (well almost) resolution:

only if want to print something bigger than 8x10

I used 5 mp, for me the real thing is the printed thing

and at 4x6 or 8x10, 3 mp is great!

in monitor ok you zooming and see more detail at 5m than 3 mp, but printed realy not see diference, for me.

thanks.

http://photos.yahoo.com/floresfrias

I did my own tests after reading your discussions. The results: It's better to use standard quality JPEG setting at 2560 than to lower the resolution to 2048 and stick with the fine JPEG quality. Both settings result in files that are about the same size (about 1.4 Meg, or 2/3 the size of the fine 2560 images).

Methodology: I put the camera on a tripod and took 4 shots, with low and high quality jpeg compression and 2048 and 2560 resolutions. The subject was a flower, and there was background imagery that would be out of focus, and therefore would tend to show some jpeg artefacts. I used f/4 to maximize sharpness. I used PanoTools "Sync 256" (Lanczos) to resize the smaller images up to the size of the larger ones. I put all 4 images into a single Photoshop document, using 4 layers. I then tried hiding and showing layers to see how the images compared.--Peter Epstein

TurboTed said:

Yossi said:

I did this experiment too a couple of months ago.
Shot in bright daylight, two images - one in 3.3 MP the other in 5
MP both with the 707 camera. The differences seen in big
enlargements are really small. Some extra background noise and a
tiny loss of resolution.
When I compared the same pictures to similar ones shot with my
Nikon CP990 I could see a much bigger difference.
Can't figure out why, but maybe I'm missing something somewhere.

Click to expand...

Click to expand...

The 990 is natively 2048, while the 707 shooting at 2048 is making a smaller image out of a whole 5Mp original. Everything is being down-sampled to create that 2048 image. So it is way bettter than a native 2048 image from any camera.

TurboTed said:

This is very interesting but I don't think you are missing
anything. It's a result of the color and luminance interpolation
inherint in traditional color CCDs. This wouldn't be the case with
Foveon's new chip.

Click to expand...

Right. Because that one has each pixel fully defined as a pure color without having to consult with pixels around it.

TurboTed said:

TurboTed said:

I do not dare to shoot at 3.3 MP from my 707 being afraid that I've
overlooked something.
I chickened out and purchased 4 more 128 MB MS cards (have now 8
altogether) and shoot at 5 MP.

Yossi

Click to expand...

Click to expand...

Do the test for yourself. Shoot 5 Mp images and 3 Mp images and stack them after making size adjustments. Blow them up on your monitor and watch the effect of switching one layer on and off. It won't take too much time to prove the effectiveness of the 2048 file to your own nose.

As for noise? When you shoot a ISO 400 image at 5 Mp, you are saving all the grain. But when you downsample the 5 Mp image to 3 Mp, the grain is being somewhat averaged, just as the detail is. The difference is this: detail in the 5 Mp image is not pixel-perfect, so effectively, the detail simply shrinks without blurring away. But the grain IS pixel-perfect, so it tends to mush together with surrounding samples and gets a bit noise-reduced.

So you end up with an image that looks more like ISO 300 but with 97% of the detail of a 5 Mp image shot at ISO 400. Not a bad trade.

When you shoot images at SXGA, 1280 pixels wide, the grain disappears completely, and the image still keeps about 67% of the detail from the full-size shot even though it is only 50% of the full image width. Now that's a pixel-perfect image.

At a 5 x 7 print size, the SXGA shot is laying 182 perfect pixels on every inch of paper. FYI, the dot screen on fine printing is less than this, and it's a screen!

-iNova

miles miller said:

from a practical standpoint ,without upsampeling does this realy
make a difference if we are just talking about minor croping and
printing at 8x10 size?

Click to expand...

Since that 2048 image contains so much of the original frame's detail, cropping can be done with the same abandon.

-iNova

Peter iNova said:

TurboTed said:

Yossi said:

I did this experiment too a couple of months ago.
Shot in bright daylight, two images - one in 3.3 MP the other in 5
MP both with the 707 camera. The differences seen in big
enlargements are really small. Some extra background noise and a
tiny loss of resolution.
When I compared the same pictures to similar ones shot with my
Nikon CP990 I could see a much bigger difference.
Can't figure out why, but maybe I'm missing something somewhere.

Click to expand...

Click to expand...

The 990 is natively 2048, while the 707 shooting at 2048 is making
a smaller image out of a whole 5Mp original. Everything is being
down-sampled to create that 2048 image. So it is way bettter than a
native 2048 image from any camera.

Peter iNova56624 said:

This is very interesting but I don't think you are missing
anything. It's a result of the color and luminance interpolation
inherint in traditional color CCDs. This wouldn't be the case with
Foveon's new chip.

Click to expand...

Right. Because that one has each pixel fully defined as a pure
color without having to consult with pixels around it.

Peter iNova56624 said:

Peter iNova56624 said:

I do not dare to shoot at 3.3 MP from my 707 being afraid that I've
overlooked something.
I chickened out and purchased 4 more 128 MB MS cards (have now 8
altogether) and shoot at 5 MP.

Yossi

Click to expand...

Click to expand...

Do the test for yourself.

Click to expand...

I did a quicker and simpler test. I took an original 5mp image (fine setting) and reduced it to 2048 in PhotoShop and then immediately resized it back to 5mp. It was ever so slightly softer, but I must admit, the detail was 99.9% there! I had to zoom in more than 200% to see any difference worth noting. Of course I went a step further and reduced the 5mp picture to 1600 and 1280 and 800 and 640, just for fun.

Peter iNova56624 said:

Shoot 5 Mp images and 3 Mp images and
stack them after making size adjustments. Blow them up on your
monitor and watch the effect of switching one layer on and off. It
won't take too much time to prove the effectiveness of the 2048
file to your own nose.

As for noise? When you shoot a ISO 400 image at 5 Mp, you are
saving all the grain. But when you downsample the 5 Mp image to 3
Mp, the grain is being somewhat averaged, just as the detail is.
The difference is this: detail in the 5 Mp image is not
pixel-perfect, so effectively, the detail simply shrinks without
blurring away. But the grain IS pixel-perfect, so it tends to mush
together with surrounding samples and gets a bit noise-reduced.

So you end up with an image that looks more like ISO 300 but with
97% of the detail of a 5 Mp image shot at ISO 400. Not a bad trade.

When you shoot images at SXGA, 1280 pixels wide, the grain
disappears completely, and the image still keeps about 67% of the
detail from the full-size shot even though it is only 50% of the
full image width. Now that's a pixel-perfect image.

At a 5 x 7 print size, the SXGA shot is laying 182 perfect pixels
on every inch of paper. FYI, the dot screen on fine printing is
less than this, and it's a screen!

-iNova

Click to expand...