F707's 5mp vs 3mp again....

[TurboTed]

As my last argument I give a hypothetical example. Imagine taking
any image and resize it 200% using no interpolation. Now you have
a large image made of 4 pixel squares where each of the 4 pixels in
each square are the same color. Assuming this to be an original
from our hypothetical camera, what would happen if you resize this
image by 50% and enlarge it by 200% again without any
interpolation. I believe you'll just end up with the same image of
2x2 pixel squares.

I like this, because it actually helps. Assume a fairly simple
reduction algorithm (in fact, you already have), in which each one
pixel is 'surviving' pixel and will be given the average value of
itself and its west, southwest, and south neighbor. These three
neighbors are then thrown out. Also assume a row of 2x2 squares
next to each other, alternating between squares with four pure red
pixels, and squares with 4 pure green pixels. If the simple
reduction algorithm starts in the northeast corner of each square,
your 50% reduction with result in a row alternating between red and
green pixels. However, if it starts in the northwest corner
(because it doesn't know you have neat 2x2 squares), you're going
to end up with a row of 128R 128G 0B pixels. Upsample that row by
200%, and your left with a 2 rows of 128R 128G 0B pixels. And that
is a simple reduction algorithm. You get more complex, and you
look to more neighbors, and then you look for patterns, etc, etc.
But, I think it would be a great test of, say, photoshop's
resampling algorithms. Create some nice grids made up of 2x2
squares in which each pixel has an identical pure color. Then
resample down and up, and see what the result is.

That is interesting, I've never considered going in a different direction. I would propose, though, that someone who wrote an algorythm to start at the northwest corner would in fact average the pixels to the east, southeast and south so to avoid averaging non-existant pixels. In this case, it would end up with alternating red/green pixels.

I don't know the math either and it goes
against common sense, but if you can imagine fractional pixels or
the fact that each pixel in our original actually contains some
information from adjacent pixels then maybe reducing these images
by some nominal amount may not really lose any info.

I say you can have fractional pixels. Take an image, use a sophisticated algorythm to scale it up 20%. You'll end up with more pixels each of which will have a value equal to its original value and some fraction of its adjacent pixels. Technically, each pixel contains a little more information than it used to, not more detail but it does contain some component of its neighbors. You now have no more information than you used to but it is taking more pixels to display it. That is what we are saying is happening when the RGBR mosaic is interpolated resulting in a camera's native image size.

You don't have fractional pixels, and when you change a pixels
value, it is changed. You might be able to apply some fancy
algorithms and do a decent job guessing what info was lost, but its
attempting to place lost info. That's why test 1 showed a softer
image.

Anyway, got to run, off to go camping with my sons!

Are you avoiding my test #2??? ;-)

Have a great time camping, I can't wait until my kids are older and we go on a camping trip.

Howdy.

 

[Shay Stephens]

A mosaic sensor has two green pixels and one red and one blue pixel. The green pixels capture the luminance, which is responsible for the resolution of the sensor. The red and blue pixels are there for color data. So a 5MP sensor only has half of it's pixels recording resolution. A 5MP camera is then equivalent to a true 2.5MP camera like a Foveon.

Now, what will happen if you let the camera downsample the image to 2.5MP and then later upsample it to 5MP. What you wind up with is a 3rd generation file after upsampling to 5MP (original image, downsampled, upsampled). You will be loosing something in the process. It's not a lot but it is detectable.

 

[Bobbo]

A mosaic sensor has two green pixels and one red and one blue
pixel. The green pixels capture the luminance, which is
responsible for the resolution of the sensor. The red and blue
pixels are there for color data. So a 5MP sensor only has half of
it's pixels recording resolution. A 5MP camera is then equivalent
to a true 2.5MP camera like a Foveon.

Now, what will happen if you let the camera downsample the image to
2.5MP and then later upsample it to 5MP. What you wind up with is
a 3rd generation file after upsampling to 5MP (original image,
downsampled, upsampled). You will be loosing something in the
process. It's not a lot but it is detectable.

Somehow the word resampling has gotten all intertwined into this subject.

The F707 takes the RAW data and creates a 5MP or a 3MP image.

The resolution of the F707 is less than 3MP.

Does the 3MP image have the same detail as the 5MP image? That's the question.

I don't advocate resampling up in normal picture taking. If the you take a 3MP picture don't resample it up to 5MP, leave it as is. The unfortunate but necessary reason the 3MP images in the examples were resampled was to have images that were the same size for visual comparison, because it is difficult to compare detail in images that are not the same size. Resampling the 3MP example added some softness, that's what happened to TurboTed's first example. If adding a little unsharp mask to the 3MP example makes the comparison similar then the above question is answered. Did you see Peter iNova's thread to my message? And Peter is not a novice.
http://www.dpreview.com/forums/read.asp?forum=1009&message=2537541

It seems lots of people are turned off by this because it defies their logic or they are not going to turn their 5MP camera into a 3MP camera. But the eyes have it. I am 95% convinced that using the 5MP mode on the F707 is about as advantageous as using the TIFF mode.

 

[Shay Stephens]

Well think of it this way, if the 3mp version had as much detail as the 5MP version, then why does it look fuzzy when resized to 5MP. There is more detail in the 5MP version. If you take the 5MP image and reduce it to the smaller size, it looks better than the 3MP image. In fact, the question then comes up, why does the image look better resampled? It seems to me that the 3MP image is not actually resampled down in the camera, it looks more like to me, upon analyzing the image, that the camera is instead using 3MP of the 5MP sensor to create the image. The red vertical barring artifacts are as strong at 3MP as it is at 5MP, and, in fact, also at 640x480. So I would say the smaller sized images from the camera are made up from fewer pixels than the 5MP version and thus have less detail.

 

[Shay Stephens]

odd worm hole there, so her is what I actually typed:

Well think of it this way, if the 3mp version had as much detail as the 5MP version, then why does it look fuzzy when resized to 5MP. There is more detail in the 5MP version. If you take the 5MP image and reduce it to the smaller size, it looks better than the 3MP image. In fact, the question then comes up, why does the image look better resampled? It seems to me that the 3MP image is not actually resampled down in the camera, it looks more like to me, upon analyzing the image, that the camera is instead using 3MP of the 5MP sensor to create the image. The red vertical barring artifacts are as strong at 3MP as it is at 5MP, and, in fact, also at 640x480. So I would say the smaller sized images from the camera are made up from fewer pixels than the 5MP version and thus have less detail.

 

[Bobbo]

I don't agree that the F707 3MP mode is not using the full chip. Look at the image from my original message on this subject where you'll also see an example from an excellent 3MP camera. The 2D subject (everything is in focus) frozen by the fast flash (no camera or subject movement) allows a better comparison of detail.

 

[Shay Stephens]

Let's use an examp[le that is more illustrative:



Note the red artifacts on the pen. They are nearly identical in the 5MP and 3MP image, but are largely gone in the resampled image. The red artifacts are a result of the 707's mosaic interpolation. So if the 3MP image were resampled in the camera it would look closer to the resampled image shown, but it looks instead more like the 5MP version which means it was created with few pixels from the CCD using the same interpolation as the 5MP image.

 

[Jimmy Chen]

If you take the 5MP image
and reduce it to the smaller size, it looks better than the 3MP
image.

And here are actual 5MP and 3MP crops, NO up-sampling or down-sampling was done in PS. The 117 series are 5MPs and the 118 series are the 3 MPs.

The "A" files are 360pX360p, to show the captured image difference when the crop is down to the same pixel count. 117 clearly uses more pixels to capture more details (smaller area captured). The B files are crop to show the same area. The 117 file is 276x272 (90k), and the 118 file is 220x218 (62k).

So either case, w/o any retouching the 5MP is capturing more detail then the 3MP.

117a



117b



118a



118b



jc

 

[Bobbo]

Let's use an examp[le that is more illustrative:



Note the red artifacts on the pen. They are nearly identical in
the 5MP and 3MP image, but are largely gone in the resampled image.
The red artifacts are a result of the 707's mosaic interpolation.
So if the 3MP image were resampled in the camera it would look
closer to the resampled image shown, but it looks instead more like
the 5MP version which means it was created with few pixels from the
CCD using the same interpolation as the 5MP image.

I don't agree with that. The 3MP image should look just like the 5MP image. The camera is creating a 2.8MP image and enlarging it to 5MP with empty magnification. It is doing the same thing at 3MP. Both 3MP and 5MP images are being created with the same data. When you resample afterwards with software the image is softened. That's why resampling will not add detail but take some away and is, as I said, unfortunate but necessary for comparison. The 3MP and 5MP images are virtually identical in your example. The F707 in camera resampling, if you can call it that, is much cleaner than post software resampling, this is also apparent with the 2X digital zoom.

 

[TurboTed]

It seems lots of people are turned off by this because it defies
their logic or they are not going to turn their 5MP camera into a
3MP camera. But the eyes have it. I am 95% convinced that using the
5MP mode on the F707 is about as advantageous as using the TIFF
mode.

Interesting analogy.

 

[TurboTed]

Well think of it this way, if the 3mp version had as much detail as
the 5MP version, then why does it look fuzzy when resized to 5MP.

It could also be one or both of the following reasons. Either we're seeing the advantage the camera has working with the RAW internal data to create the 5mp image, or the PhotoShop algorithm I used is not as aggressive with the sharpening.

Do you still think the two images in Test #2 look any different now? Which one is the 5mp image this time?

There is more detail in the 5MP version. If you take the 5MP image
and reduce it to the smaller size, it looks better than the 3MP
image. In fact, the question then comes up, why does the image
look better resampled? It seems to me that the 3MP image is not
actually resampled down in the camera, it looks more like to me,
upon analyzing the image, that the camera is instead using 3MP of
the 5MP sensor to create the image. The red vertical barring
artifacts are as strong at 3MP as it is at 5MP, and, in fact, also
at 640x480. So I would say the smaller sized images from the
camera are made up from fewer pixels than the 5MP version and thus
have less detail.

 

[Shay Stephens]

This is a comparison, the smaller crop is from the resampled 5MP image, the larger crop is the native 3MP image:



The 5MP resampled to 3MP looks much better than the native 3MP. If the F707 had empty enlargement, then resampling in PS would not make the photo look better than the native 3MP image. You can't get something from nothing, and it is the case here too. The reason the 5MP image resampled to 3MP looks better than the native 3MP image is because there is more actual image data in the 5MP image as opposed to the 3MP image.

 

[Bobbo]

Let's use an examp[le that is more illustrative:



Note the red artifacts on the pen. They are nearly identical in
the 5MP and 3MP image, but are largely gone in the resampled image.
The red artifacts are a result of the 707's mosaic interpolation.
So if the 3MP image were resampled in the camera it would look
closer to the resampled image shown, but it looks instead more like
the 5MP version which means it was created with few pixels from the
CCD using the same interpolation as the 5MP image.

This is an excellent example of the destructive powers of resample. Look at the short vertical lines at the interface of the red pen and the blue and black pens, these are artifacts, they are not real features of the subject, but the software that resampled the 5MP image to 3MP didn't know that. These artifacts became real details of the image and the resample destroyed them, but the 3MP camera image preserves them perfectly. That's what I've been saying, the 5MP and 3MP images appear to be similar.

 

[Shay Stephens]

Image one is the 3MP and image 2 is the 5MP but they do look much closer now having had the unsharp mask applied.

 

[Shay Stephens]

But my point is that they should not look the same, the fact that they are means to me that the camera is using fewer pixels to render the 3MP image. I wish there were a way of knowing how they achieve the 3MP image, it would shed a lot of light on this. But my gut is telling me that fewer pixels are being used, and thus the resolution is reduced.

 

[Bobbo]

But my point is that they should not look the same, the fact that
they are means to me that the camera is using fewer pixels to
render the 3MP image. I wish there were a way of knowing how they
achieve the 3MP image, it would shed a lot of light on this. But
my gut is telling me that fewer pixels are being used, and thus the
resolution is reduced.

I'm saying the 5MP and 3MP images look similar and that implies they have the same detail. Your saying not so, to have the same detail they shouldn't look the same. I don't follow your logic. Your impressed with the resampled 3MP image, you say it looks better than the 3MP camera image. Better than what? Ideally the resampled image would retain the detail of the original, it removed the fine detail and you say that's good. Again, I believe the camera is creating the 3MP and 5MP images from the same data base, that's why they look similar.

 

[scorch]

I haven't followed this whole thread, and don't understand how a 5mb camera's 5mb image isn't any better than it's 3mb image. Does this mean that 3mb cameras 3mb images are no better than say it's 2mb images?

The other theory seems to be, how does a 5mb camera create 3mb images? Seems there are 3 possible ways.
1. Use the whole CCD and throw out pixels
2. Use the whole CCD and no pixels are thrown out
3. Use only part of the CCD.

Number one seems the most likely to me. But that would suggest that creating an even smaller image, such as 800 x 600 would be worse than using a good 1mb camera.

Number 2 would mean that each pixel on the CCD is resizable depending on the the resolution selected. That seems unlikely to me.

Number 3 would mean that pixels are thrown out before they reach the CCD. That seems unlikely to me also.

Of course, I could be missing this point also.

Ken

 

[Shay Stephens]

Well if you like the red artifacts, how about the noise? When resampling down the 5MP in a photo editor you also reduce the noise in the image:

 

[Bobbo]

Well if you like the red artifacts, how about the noise? When
resampling down the 5MP in a photo editor you also reduce the noise
in the image:

Again, resampling changes the characteristics of the image, removing not adding detail. Once you have an image file, software image editors have no way of knowing if features are noise, artifacts or actual details. If the noise in this image was real surface texture details resampling would have also reduced them. If you resample the 3MP camera image by 80% you will get similar results, which doesn't prove anything. As I stated in another thread "Somehow the word resampling has gotten all intertwined into this subject". Resampling has nothing to do with this subject.

 

[Howdy]

OK, TurboTed you badgered me into it. Actually, I finally had a moment to spare. Figures that I don't have that moment until nearly midnight on Saturday.

OK, my first impression is that Image 1 seems to have several distinct sharpening artifacts.

Image 2 has a better 'feel.' Detail seems to arise from detail, ie expected differences between neighbouring pixels, not artificial differences.

So, my vote would be Image 1 started life as 3mp, then was resized to 5mp and then sharpened (oversharpened?)

Image 2 came from camera as 5mp image.

Howdy