Ad
Ad
Ad
Pages: « 1 2 3 [4] 5 »   Bottom of Page
Print
Author Topic: SIGMA DP-1  (Read 30833 times)
Petia
Newbie
*
Offline Offline

Posts: 17


« Reply #60 on: February 18, 2008, 10:27:57 AM »
ReplyReply

Quote
I can see the aliasing with the images downsampled to fill my monitor.  That's all I need to have a problem.

In these conditions there's no discussion possible. The artifacts are in your head.
Logged
Graeme Nattress
Sr. Member
****
Offline Offline

Posts: 582



WWW
« Reply #61 on: February 18, 2008, 12:06:44 PM »
ReplyReply

As far as I've seen, the artifacts are in just about every in-focus Sigma image I've seen when viewed 1:1.

Graeme
Logged

www.nattress.com - Plugins for Final Cut Pro and Color
www.red.com - Digital Cinema Cameras
John Sheehy
Sr. Member
****
Offline Offline

Posts: 838


« Reply #62 on: February 18, 2008, 12:18:15 PM »
ReplyReply

Quote
As far as I've seen, the artifacts are in just about every in-focus Sigma image I've seen when viewed 1:1.
[a href=\"index.php?act=findpost&pid=175712\"][{POST_SNAPBACK}][/a]

And 1:1 with 4.6M pixels is not that far away from full-screen scaling with hires monitors.
Logged
John Sheehy
Sr. Member
****
Offline Offline

Posts: 838


« Reply #63 on: February 18, 2008, 12:42:41 PM »
ReplyReply

Quote
In these conditions there's no discussion possible. The artifacts are in your head.
[a href=\"index.php?act=findpost&pid=175683\"][{POST_SNAPBACK}][/a]

I'll continue anyway, without you.  The main problem I have with aliased capture is the lack of sub-pixel precision, or the "snap-to-grid" effect.  This does not improve tremendously with minor ratios of downsampling, because even though the downsampling increases sub-resultant-pixel precision, the degrees of freedom are still restrained.  I estimate about 25%-30% linear resizing is necessary so that the snap-to-grid effects are lost.  My monitor has a height of 1050 pixels.  The Sigma images are 1760 pixels high.  That's a downsampling of 1050/1760, or 60%.

Back to one of the points you made earlier; the issue of blurring before or after capture:

1)  You are overlooking the very pertinent fact that cameras without AA filters have blind space between the pixels; cameras with AA filters do not.  This, in itself is one of the major contributors to aliasing.  A true box capture with no spatial waste aliases far less than even one that captures 90% of the area effectively.  Two diagonal-neighbor pixels can easily miss a thin line passing through them completely.  Even if the area that is totally blind is very small, the fringes can be less sensitive and still have more weighting towards the middle, so even with microlenses, vestiges of point-like sampling can still be present.

2) Blurring after capture can not reduce low-frequency aliasing effects like picket fences that change in luminance over their run.

I am currently working out a method of doing simulations of the issues we are discussing in PS, including the dead space issue.
Logged
Petia
Newbie
*
Offline Offline

Posts: 17


« Reply #64 on: February 18, 2008, 02:26:24 PM »
ReplyReply

Quote
As far as I've seen, the artifacts are in just about every in-focus Sigma image I've seen when viewed 1:1.

Graeme
[a href=\"index.php?act=findpost&pid=175712\"][{POST_SNAPBACK}][/a]


Right. What I'm saying is that precisely I don't view image crops at 100%. I view them in prints. At normal distance. Without a magnifying glass.
I don't expect my setup to give me 60 lp/mm. I don't own a 10k$ Leica. This thread is about a $700 compact camera (for those who forgot).
Besides, some people here should understand that setting parameters in the image processing pipeline goes beyond chossing a jpeg compression level (personally I never shoot jpeg). From the raw cmos signal to the image, there are a bunch of transformations, and in all cameras I know you can set your personal sharpening level. Those who complain about jaggies should first check whether this is the solution to their problem (if it is real). In any case if you've to print a giant poster, a 0.5 Gaussian blur in Photoshop will be quite efficient without affecting much the overall sharpness.
When the Kodak DCS 14n appeared without anti-aliasing filter, pixel experts were duly horrified (" 't was a bold but stupid move"). When the Nikon D70 was released with an OLPF that was in the high cutoff range, example of situations where moire appears flourished on the web (and this is far worse than jaggies, because the artifactual color is really apparent at the scale of > 10 pixels. To own one, I know this is *really* a marginal problem (the small and moderately bright viewfinder is far more annoying, but that's not visible on a website image viewed at 1:1...). When the Leica M8 appeared, well I dunno what our expert pixel peepers thought (probably "heck, no OLPF, but I can't really say on a forum that Leica engineers are clowns, so let's forget about it").
Now it's the turn of the Foveon X3 and its derivatives. Come on guys, please talk about color rendition, color profiling, ability to render skin tones, highlight details, max sensitivity, color and luminance noise, whatever, but let the 100% crop fantasms to those that never take pictures (or only photograph resolution charts), and consequently never show them.
Logged
BryanHansel
Sr. Member
****
Offline Offline

Posts: 369


WWW
« Reply #65 on: February 19, 2008, 11:26:21 AM »
ReplyReply

Quote
Indeed, think of a fixed 35 mm VR f2.8 lens, weather sealed compact body on top of a Nikon D80 or Canon 400D sensor... with RAW capture.

Sweet thought! I'd buy one instantly for 600 US$.

Ditto. The only other thing I'd look for is a 85mm VR included.
Logged

John Sheehy
Sr. Member
****
Offline Offline

Posts: 838


« Reply #66 on: February 20, 2008, 10:41:12 PM »
ReplyReply

Quote
I am currently working out a method of doing simulations of the issues we are discussing in PS, including the dead space issue.
[a href=\"index.php?act=findpost&pid=175732\"][{POST_SNAPBACK}][/a]

Here we go:



The basic idea here is that I started with an original with 16x as many pixels, to simulate an analog original, with a very sharp lens with low diffraction and sharp-edged subjects (although all shapes were placed in PS with anti-aliasing enabled).  All are immediately-rasterized shapes on a single layer in PS, except the strip on the right of variable frequency, which I generated mathematically in Filter Factory.  The appended "AA" means that a gaussian blur equal to 1/2 the resulting pixel width was applied before sampling.  "Box" means the sampling of each 4x4 tile of the original included all 16 pixels in the output pixel, "Vpoint" means that a 3x3 tile only was used, and there was consequently some dead space in the sampling.

Hopefully, you will see here what it is that I object to in aliased sampling.

To my eyes, in order of decreasing sampling quality they are:

1) BoxAA -best but purely theoretical, unless a camera were to use binning of extreme pixel density with an AA strength calculated for the resulting pixel.  Requires no dead or insensitive space between pixels.

2) VpointAA - similar to what we would get with a greyscale camera with an AA filter.

3) Box - theoretical unfiltered capture, with no dead  or low-sensitivity space.

4) Vpoint - Similar to what we are actually getting with Foveon and greyscale cameras with no AA filter.

The main problem with the unfiltered captures is that they have spatial jitter.  The objection that this is only visible with pixel peeping is ridiculous, because the jitter is visible with any amount of blurring, downsampling, or stepping back from the monitor unless such are so extreme that you can't recognize these captured shapes at all.  Any attempt to blur (or downsample, or step back from) the unfiltered captures after the fact makes them much softer than the filtered ones, without removing the jitter, and makes them totally incomprehensible if done to the extent that the jitter is no longer noticeable.

Now, it might take these worst-case B&W originals for you to see the jitter, but *I* see it in regular images with lower-contrast subjects, and hopefully you now have an idea of what it is I see in aliased captures that I do not like - false detail that arbitrarily says something which has nothing to do with what is really there.

YES - it lets you know that you have successfully focused on the medium.

NO - it is not the real subject.
Logged
John Sheehy
Sr. Member
****
Offline Offline

Posts: 838


« Reply #67 on: February 21, 2008, 08:15:13 AM »
ReplyReply

Quote
Here we go:[a href=\"index.php?act=findpost&pid=176335\"][{POST_SNAPBACK}][/a]

Take special note of the strip on the lower right; the original increases in line frequency going down; the low-frequency artifacts from nyquist mirroring in some are there forever, with no chance of removal.
Logged
Petia
Newbie
*
Offline Offline

Posts: 17


« Reply #68 on: February 21, 2008, 09:02:01 AM »
ReplyReply

It's a very nice demonstration, and shows the aliasing artifacts quite nicely, but it would be more convincing if you didn't assume that the lens resolved 250 lp/mm on a Foveon-type (20 mm wide) sensor.
The day such a lens will be manufactured, I'm not sure you and me will be alive (if it ever happens)...
Logged
Graeme Nattress
Sr. Member
****
Offline Offline

Posts: 582



WWW
« Reply #69 on: February 21, 2008, 09:28:07 AM »
ReplyReply

The Foveon in the SD14 is 20.7mm wide, and has 2640 pixels, therefore it's resolution is 127p/mm or about 63 lp/mm.

Graeme
Logged

www.nattress.com - Plugins for Final Cut Pro and Color
www.red.com - Digital Cinema Cameras
John Sheehy
Sr. Member
****
Offline Offline

Posts: 838


« Reply #70 on: February 21, 2008, 12:28:42 PM »
ReplyReply

Quote
The Foveon in the SD14 is 20.7mm wide, and has 2640 pixels, therefore it's resolution is 127p/mm or about 63 lp/mm.
[a href=\"index.php?act=findpost&pid=176403\"][{POST_SNAPBACK}][/a]

Shift everything by 1/2 pixel, and it won't resolve it at all, though.  0 contrast.
Logged
John Sheehy
Sr. Member
****
Offline Offline

Posts: 838


« Reply #71 on: February 21, 2008, 07:53:44 PM »
ReplyReply

Quote
Shift everything by 1/2 pixel, and it won't resolve it at all, though.  0 contrast.
[a href=\"index.php?act=findpost&pid=176450\"][{POST_SNAPBACK}][/a]

In my examinations of luminance resolution tests amongst cameras, I've found that both bayer cameras and Foveons have a consistent, reliable maximum resolution (the only kind there really is, IMO; relying on luck is not "resolution" in my mind), is about the same, relative to the pixel frequency.
« Last Edit: February 21, 2008, 07:54:04 PM by John Sheehy » Logged
John Sheehy
Sr. Member
****
Offline Offline

Posts: 838


« Reply #72 on: February 21, 2008, 07:55:58 PM »
ReplyReply

Quote
It's a very nice demonstration, and shows the aliasing artifacts quite nicely, but it would be more convincing if you didn't assume that the lens resolved 250 lp/mm on a Foveon-type (20 mm wide) sensor.
The day such a lens will be manufactured, I'm not sure you and me will be alive (if it ever happens)...
[a href=\"index.php?act=findpost&pid=176399\"][{POST_SNAPBACK}][/a]

... but I see these same artifacts in actual Sigma images; mortar periodically missing, etc, etc.
Logged
Petia
Newbie
*
Offline Offline

Posts: 17


« Reply #73 on: February 22, 2008, 07:16:23 AM »
ReplyReply

Quote
The Foveon in the SD14 is 20.7mm wide, and has 2640 pixels, therefore it's resolution is 127p/mm or about 63 lp/mm.

Graeme
[a href=\"index.php?act=findpost&pid=176403\"][{POST_SNAPBACK}][/a]

And if you assume that the original image is 4x more defined that the sensor grid, as John did, you get 4x63 = 252 lp/mm.
« Last Edit: February 22, 2008, 07:18:22 AM by Petia » Logged
John Sheehy
Sr. Member
****
Offline Offline

Posts: 838


« Reply #74 on: February 22, 2008, 10:42:08 AM »
ReplyReply

Quote
And if you assume that the original image is 4x more defined that the sensor grid, as John did, you get 4x63 = 252 lp/mm.
[a href=\"index.php?act=findpost&pid=176631\"][{POST_SNAPBACK}][/a]

There are plenty of lenses that do that in their sweet spots with at least traces of contrast, and as I said, all my original shapes were placed with anti-aliasing.  The only thing that isn't anti-aliased is the very bottom of the variable-frequency strip, as I did not over-sample it in the calculations.  It is aliased enough to have some spatial jitter, but no nyquist mirroring is visible (it increases in frequency very slightly up until the end).

Regardless, the bottom line is that these types of artifacts are visible in real Sigma images with real lenses.
« Last Edit: February 22, 2008, 10:43:01 AM by John Sheehy » Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8853


« Reply #75 on: February 22, 2008, 11:21:29 AM »
ReplyReply

Whilst I understand the concerns about alaising artifacts and false detail from a viewpoint of total accuracy, it seems we have a phenomenon here which could legitimately appeal to personal taste.

If you are attempting to create a work of art, then you manipulate the image to get the effect that satisfies you, and you hope it also satifies the customer if you are attempting to sell the work.

It would seem clear that John Sheehy would not be a potential customer for such a work produced from a Sigma camera because he is so finely tuned to the effects of aliasing artifacts, which he doesn't like, and which I think he would notice and identify as such where most people wouldn't.

Most people might be deceived into thinking that the Sigma image is just plain sharp and crispy. Perhaps they don't care if some of that detail is false. Perhaps here, with the characteristics of the Foveon sensor, we have a type of photographic impressionism; the sharpness of the photograph caricaturised.
Logged
John Sheehy
Sr. Member
****
Offline Offline

Posts: 838


« Reply #76 on: February 22, 2008, 03:24:34 PM »
ReplyReply

Quote
Most people might be deceived into thinking that the Sigma image is just plain sharp and crispy. Perhaps they don't care if some of that detail is false. Perhaps here, with the characteristics of the Foveon sensor, we have a type of photographic impressionism; the sharpness of the photograph caricaturised.
[a href=\"index.php?act=findpost&pid=176671\"][{POST_SNAPBACK}][/a]

You could look at it that way, but I don't think that this is what the typical person impressed by the spatial aspects of Sigma images sees; I think that they assume that they are really seeing better resolution than what is really there.

Do you really want impressionism to happen at the capture level?  You can do much more with an image aliasing it or mosaicing it yourself, after the fact of capture.

I'm looking towards the future, and it scares me that public demand (which drives marketing to some degree) seems to be for low pixel densities with pseudo-detail.  Much more can be done with a higher-resolution sensor.  Software aliasing filters would be easy to implement, with all kinds of degrees of freedom; a filter could present a square to you in the middle of its dialogue, and you just draw a luminance mask inside and outside of it to show the weighted influence of the source area to the output pixels.  I don't need any camera capturing RAW for me with artificial texture.
Logged
Petia
Newbie
*
Offline Offline

Posts: 17


« Reply #77 on: February 27, 2008, 02:32:27 PM »
ReplyReply

Quote
I'm looking towards the future, and it scares me that public demand (which drives marketing to some degree) seems to be for low pixel densities with pseudo-detail.  Much more can be done with a higher-resolution sensor. 

Your scare is very weird, because it's not at all backed by the facts.

What scares me is exactly the contrary: artificial demand for more resolution and more megapixels (the more the better, isn't it? - marketing blurb). Once upon a time a company called Fuji made a series a fine compact cameras around a 6.3 MP sensor (F10, F11, F20, F30). These were the first compact cameras that were able to produce decent images at 800 ISO. They sadly were the last too. Fuji dropped this in under the pressure of the pixel-peeping crowds that always demand more pixels. The low-light capable F30 was replaced by a 10 MP F50 which is no better than the rest of the 10-12 MP compact cameras: useless over 400 ISO.

For the Foveon as for the other CMOS/CCD sensors (and as for silver halide emulsions...), there's a  trade-off between resolution and sensitivity. It's a good thing Foveon did not choose to increase resolution, because noise would be even more problematic. It is problematic enough to my taste.
I clearly prefer crisp 8x12" prints (or slideshows) with low noise to accurate 1:1 luminance detail on my monitor and useless prints plagued with color noise.

But John, you can rest assured that the vast majority still believes it's normal to have Christmas trees in any forest shot at 800 ISO. And that you need to have enough resolution for 20x30" prints even if the lens resolves 30 lp/mm (24x36mm-equivalent) and less than thousand photographers on earth print a 20x30" each year.
Logged
Petia
Newbie
*
Offline Offline

Posts: 17


« Reply #78 on: February 27, 2008, 02:47:51 PM »
ReplyReply

Quote
Most people might be deceived into thinking that the Sigma image is just plain sharp and crispy. Perhaps they don't care if some of that detail is false. Perhaps here, with the characteristics of the Foveon sensor, we have a type of photographic impressionism; the sharpness of the photograph caricaturised.

Exactly.
As well as most people will be deceived into thinking that the colors of an image derived from a Bayer array sensor are the original ones. In some way the sharpness of a Bayer-type image is also a caricature as it is reconstructed by filtering (unsharp masking etc - which in some situations lead to sharpening artifacts).

All these technologies rely on what kind of approximation of the original light stimulus we are ready to accept as a faithful representation.
Logged
Graeme Nattress
Sr. Member
****
Offline Offline

Posts: 582



WWW
« Reply #79 on: February 27, 2008, 02:50:05 PM »
ReplyReply

Because silicon is a poor colour filter, the colours on a Foveon are somewhat guessed too....

Graeme
Logged

www.nattress.com - Plugins for Final Cut Pro and Color
www.red.com - Digital Cinema Cameras
Pages: « 1 2 3 [4] 5 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad