A.pellucida - Blue Filter

I have been doing some shots of AP using a Coolpix 4500 and a blue interference filter (436 nm). My initial pictures were in colour mode (RGB) and were awful. Breaking them down into RGB led to an entirely white blue component and unhelpful R and G components.

However, setting the camera for monochrome gives quite decent pictures which reveal the additional resolution expected. Presumably this has something to do with how such cameras deal with spectrum components but it is not obvious why the monochrome option is so markedly better. Can someone explain in non-technical terms please? Also, would the same principle apply to a green filter (though I haven't had any problem in colour mode with such)?

Selwyn

Selwyn,
I too would really like to hear an explanation re consumer camera monochrome mode.

If you look at the basic Bayer filter (applicable to Coolpix cameras) the blue component information is 'recorded' as one of four pixels within this filter array. Therefore, inherently, blue information alone should lead to a 2x degradation in resolution.

Maybe other members will pick it up from there....

Regards, Keith

"A. S. St Leger" wrote:
> ...how such cameras deal with spectrum components but it is not
> obvious why the monochrome option is so markedly better. Can someone
> explain in non-technical terms please?

"A. S. St Leger" wrote:
> I have been doing some shots of AP using a Coolpix 4500
> and a blue interference filter (436 nm). My initial pictures were in colour mode (RGB) and were awful.

It might sound odd, but check your exposure times. I suspect that your colour mode shot used a long exposure time that caused all the blue-sensitive photosites to saturate, while leaving the red and green photosites to just accumulate noise.

As to resolution, the issue that Keith mentions is real ("blue information alone should lead to a 2x degradation in resolution"), BUT at high magnification it's quite possible that your camera has more than enough photosites to catch all the detail that's in the optical image, even though only 1/4 of them are sensitive to blue.

We would have to know a lot more about your setup to calculate it, but here's a quick and dirty empirical test.
Shoot an image with the camera set to its largest image size. Pull the image into Photoshop and duplicate it. Resize the duplicate to be half as many pixels wide and high, then resize it back to the original dimensions. Compare the resulting image with the original.
If the process of shrink & expand by 2X does not cause much detail to disappear, then you know that using only 1/4 of the pixels is still enough.

-Rik

Selwyn,
I think this must have something to do with how the camera processes the image after it captures it.

When you think about it, all digital photographs are basically "monochrome," since the output of the chip is electrons and not photons. You could probably talk about different "colors" (wavelengths) of electrons, but nobody has figured out how to make an imaging chip do that kind of thing.

In a Bayer mask, think of a group of 4 pixels on a chip. One pixel will have a blue filter over it, one pixel will have a red filter, and two pixels will have green filters. That pattern of 1 blue : 1 red : 2 green is repeated over the entire chip. The camera reads the output of each pixel and assigns "blue" to those pixels with blue filters on them, "red" to the pixels that have red filters over them and "green" to the ones with green filters.

I can see how the resolution would go to pot with a blue filter in front of the chip. Only the blue pixels would register any photons and give any kind of output (electrons) for the camera to process. In fact, the resolution would only be 1/4 of what the chip was capable of (only 1/4 of the pixels detect blue light).

In the monochrome mode, maybe the camera's onboard software is smart enough to "know" it's only forming a picture from the blue pixels and decides, "aha, only blue pixels are registering. I'd better assign this signal to every pixel on the chip, interpolate them (make a guess at the intensity values for the non-registering pixels) and form a black-and-white image"!

That's purely a guess, but with a blue filter of a specified wavelength in front of the chip, there's no way the red and green pixels are going to see anything. So for a decent resolution monochrome image, it almost certainly is a result of some kind of interpolation going on in the camera after the raw image is captured.

My two cents' worth!
Cheers, Bob Chiovetti

Thank you everybody for responding. I shall take up the point of checking the exposure. Increasingly I am using monochrome when attempting to discern fine detail and find that ImageJ handles monochrome more flexibly than RGB at present.

When I was referring to increased resolution I meant of structures identifiable in the image rather than the pixel content of the image. The images were taken at 4x optical camera zoom so this seems to give sufficient density of pixels for my purpose.

You may care to look at images 12 and 13 in the album below. Image 13 was kindly enhanced for me by Don Williams. These show more detail in AP than I have yet been able to achieve with broad spectrum light. You need to use the magnify option for the images and then scroll to the lower end of the specimen.
  Link : http://picasaweb.google.com/...

Selwyn

Hi Selwyn,
Here is the response of the color chip on your camera:
 Link : http://www.vaytek.com/specDVC.htm
You can see that at 446 nm it exposes the green to some degree as well. I expect the camera my each that to some degree weighting the blue pixels more but slit getting exposure information from the green pixels.

Hi Bob,
You point out why consumer camera are not good choices for anything but color photography, If the blue filter is a narrow bandwidth, or a blue LED or sharp cut off low pass filter that in the short end of the blue spectra your 1/4 of the pixels is right.

But the other pixels come in to play more for Selwyn as his filter is well in the range the green pixels can detect. Look at the spectral response curves of Bayer filter at: http://www.vaytek.com/specDVC.htm (see link above). and the curve for blue filter used to just cool the light of a microscope to improve the contrast by taking out red light. I found a Koop curve set for the common blue glass filters:
  Link : www.besoptics.com/html/body_kopp_blue_filters.html
In the lower right are the curves for Kodak Wratten filters for converting daylight to tungsten film. The 80 is a common curve. Usually and 80A or 80B:
  Link : www.schneideroptics.com/pdfs/filters/FilterTransmissionCurves.pdf
Here are curves for most of the Minus Red Kodak Color correction filters:
  Link : www.olympusmicro.com/primer/photomicrography/ccfilters/ccred.html
The Minus Red Kodak Color correction filters might have a real place with a Bayer filter with white LEDs for monochrome images. Reducing the blue at the expense of the green should give very even exposure over 3/4s the pixels. A special routine might have to be written to fill the unexposed red pixel value with some average of the 4 or 8 joining blue and green pixel values in the raw color image an do your own conversion to monochrome.

Gordon

Hi Gordon,
Thanks for that, it is illuminating.
Obviously, as is consistent with the quality of photos made, I am getting more than 1/4 of the pixels when using the filter; the graphs make that clear.

Selwyn

Gordon and Selwyn,
OK, I'm confused, and I wonder what you guys are looking at.

http://www.vaytek.com/specDVC.htm describes RGB filters.
But http://www.dpreview.com/reviews/specs/Nikon/nikon_cp4500.asp says that Selwyn's camera uses a CYGM filter array. That's red/green/blue versus cyan/yellow/green/magenta -- those graphs don't apply to that camera.

As to the quality of photos made, the images that I see at http://picasaweb.google.com/selwyn.stleger/Amphipleura would be completely consistent with only 1 in 4 photosites being active. That's because the camera's sensor is 1704 x 2272, but what's displayed is only 738×1194 -- so every pixel that's displayed derives from 4 pixels (photosites) on the sensor. CYGM has at least 2 out of every 4 photosites that are blue-sensitive, but they aren't needed to explain the resolution shown at picasaweb.

Have I missed something here?

-Rik

Rik,
I shall have to rely on Gordon to answer that. My understanding of camera technology is basic.

Whatever the actual number of pixels that were responding to the blue light the picture serves its purpose, though as I mentioned I did use four times optical zoom.

Selwyn

Hi Rik,
I'm not sure that the dpreview information is correct. But Gordon may have input as he has been around these Coolpix cameras for quite some time. Also, DavidS tracks CCD sensor specifications.

It seems that the Coolpix 4500 uses the Sony ICX406AQ CCD sensor, which has a Bayer RGB filter array - see:
  Link : www.kscitech.com/MicGroup/Misc/Sony252.pdf

However, it also seems that this 3.9 mpixel CCD device was a higher density (photosites/pixels) version of the 3.4 mpixel ICX252AQ, whereby the square pixel size was reduced from 3.45 um to 3.12 um. Both devices have the same diagonal specification, namely 1/1.8"

Now, to add to the possible confusion, the ICX252AQ may have been used in earlier versions of the Coolpix camera and it was indeed available in both RGB and CYGM versions.

That's as far as I got!

Regards, Keith

"Rik Littlefield" wrote:
> OK, I'm confused, and I wonder what you guys are looking at.
> http://www.vaytek.com/specDVC.htm describes RGB filters. But
> http://www.dpreview.com/reviews/specs/Nikon/nikon_cp4500.asp
> says that Selwyn's camera uses a CYGM filter array.

There are actually 2 datsheets:
  Link : www.kscitech.com/MicGroup/Misc/Sony406.pdf
along with that below.

Now for another cup of black coffee....
-Keith

"scitech200" wrote:
> http://www.kscitech.com/MicGroup/Misc/Sony252.pdf

Selwyn,
I'm interested in trying a calculation re image pixel size (Sony CCD sensor spec) versus A.p feature dimensions, with the 100x objective. So a couple of questions:
- Did you capture at the max. resolution (2272x1704) or some other resolution?
- Is there any resampling involved with the cropped, published image?

However, on thinking about this we don't really know what Picasa is doing with the image that you submit - for display and download. So, I have to ask - do you use FTP? If you would be amenable, we could use FTP to transfer the original captured image for analysis.

Thanks for sharing your images with us. This is a worthwhile discussion as, up to this time, the group has only superficially adddressed the consumer digital camera 'monochrome' mode issue. The comparison with a bonafide monochrome camera is obvious, once the basic Bayer filter and corresponding pixel data processing is understood.

Regards, Keith

"A. S. St Leger" wrote:
> Whatever the actual number of pixels that were responding to the
> blue light the picture serves its purpose,...

"A. S. St Leger" wrote:
> Whatever the actual number of pixels that were responding
> to the blue ligt the picture serves its purpose,
> though as I mentioned I did use four times optical zoom.

Selwyn,
The pictures are very promising. I'm just a bit concerned, on closer study this morning, that some of the finest "details" are actually artifacts introduced by the camera.

I'm not worried about the rows of dots -- they seem to span enough pixels that everything ought to average out. But on the central "ridge" of the diatom (sorry, I don't know the proper term), there are what appear to be fine striations, diagonal with respect to the diatom. I can't help noticing that those striations seem perfectly aligned along columns of pixels. They're also quite small compared to the diatom's rows of dots (which are supposed to be difficult to image), and the striations disappear almost entirely when I do the shrink-by-2X-then-expand-by-2X test. The combination of all these aspects is worrisome.

It might be revealing to take a series of pictures, keeping everything the same *except* for rotating the camera by say 1/16 turn between shots. If the striations continue to look the same with respect to the diatom, then they are probably really present in the optical image (though they could still be some interference effect rather than directly representing striated structures). On the other hand, if the striations do something like persist in being column- or row-aligned instead of rotating smoothly to track the diatom, then that would suggest they're an artifact of some sort.

Sorry if I sound like a worry-wart. It's just that I've seen too many imaging systems do weird stuff on the edge of their design envelope, and even at 4X optical zoom, it looks like you're still pretty close to the edge with that camera. The rotation test might give a lot of information pretty quickly.

-Rik

Rik,
I much appreciate your comments. It's nice to know that the dots are convincing though I have a long way to go (if ever) before I produce one of those pictures of a whole AP with dots displayed along its length.

The items that worry you (don't worry me too much as I am delighted with the dots) might well be artifacts. I think the ridge is called a raphe (someone correct me). I'm not aware of structure there that is conceivably visible with the light 'scope but someone may know better. I think your diagnosis might well be correct. I have a second photo with the diatom displaced laterally which might also displace it slightly vertically with respect to the pixels. I can send you that and the original, 879 KB and 100Kb Jpeg images respectively. Maybe these are not too big for email in which case please tell me what address to use. Failing that we might find some other form of transfer.

I will do your suggested experiment of rotating the stage. However, I am separated from my 'scope at present and shall not be reunited until two weeks have passed. Obviously I can't guarantee to find the same specimen in the extensive strew but we shall see what happens.

Selwyn

Keith,
I'm glad this topic has aroused interest.

I don't use the highest camera mode but the next best which gives 1600x1200 (24 BPP) pixels. I tend to the view that so much of these photos is repetitive data that Jpeg compression looses little if any information. Perhaps, when pushing the camera to its limits I should use the highest mode with all its consequent inconvenience.

I hope I made it clear that the picture is the mean of nine so noise has been reduced to a third.

The published image is not enhanced in any way (other than camera defaults, many of which I have switched off) and not resampled other than what Picasa might have done. The original image from which that was cropped is only 100 KB and should easily be receivable by email if you give me a suitable address. I have a similar image which is 879 MB (I don't know why the disparity in size) taken with the diatom displaced laterally and not quite so good but probably worthy of enhancement. I don't know whether ImageJ did any automatic resampling or further compression when the stack was created or when the stack average was calculated.

I'll do FTP if you tell me how. Maybe my son could put the images in a file on his Linux system and I could give you the address.

Selwyn

Selwyn,I'm assuming you are using WindowsXP, so could you please try a temporary user FTP access at: ftp://temp@ftp.kscitech.com using Internet Explorer. You should get a popup window that requests a password. If this works OK, I'll email you a password. Then you can copy any file to the folder on the web server. There is no doubt that this is the most reliable way to transfer large files.

I'll be back to you with some thoughts on your Picasa Ap images.

Best regards, Keith

"A. S. St Leger" wrote:
> I'll do FTP if you tell me how....

Keith,
Your link causes a pop up box to appear. Please email me a temporary username and password.

Also while I am at it are there any other pictures from the Picasa set you want? Feel free to use them as you wish.

Selwyn

Selwyn,
Great, thanks again for making the images available.

I'll study the reference Picasa images and make a specific request via email, along with the FTP password.

Best regards, Keith

"A. S. St Leger" wrote:
> I can send you that and the original, 879 KB and 100Kb Jpeg images respectively.

Selwyn,
Thanks for sending me the images by email. I received AVG_Stack2.jpg and AVG_Stack3.jpg, both at size 1600x1200 pixels.

One thing that does strike me as a potential wrinkle is that 1600x1200 does not divide evenly into your camera's sensor array size, which dpreview lists as 2272x1704.

No matter what kind of Bayer filter the camera uses, it's pretty much guaranteed that the photosites will be distributed as some repeating pattern of 2x2 tiles, each tile containing the same number of blue- sensitive photosites. (The number of blue-sensitive sites could be anything from 1 to 4, depending on filter characteristics, but it should be the same in each 2x2 tile.)

Because of this, the most artifact-free reduction would be exactly 2X on each axis -- to 1136x852 from 2272x1704. Reducing to 1600x1200 means that some of the result pixels have to be derived differently from others -- either from more photosites, or using different coefficients, or being shifted strangely.

If its feasible for your setup, I recommend to have your camera shoot images in its native sensor size. There's less to go wrong that way. Then do any required reductions in postprocessing, say in Photoshop, where you can at least see how the result compares to the original.

> I will do your suggested experiment of rotating the stage. However, I am separated > from my 'scope at present and shall not be reunited until two weeks have passed. > Obviously I can't guarantee to find the same specimen in the extensive strew but we shall > see what happens.

Actually I want to rotate the camera, not the stage. The purpose of the experiment is to see how the camera captures exactly the same optical image in different orientations. Rotating the stage could change the optical image -- say if your illumination is even a little bit asymmetric. That could cause confusing results.

-Rik

Rik,
If your using a Coolpix 4500 or 995 using the lens at its widest setting reduces the artifacts lot. See:
  Link : http://www.couger.com/microscope/shootout/shootout.html

Gordon

Rik,
I take your point. I shall rotate the camera.

Selwyn

"A. S. St Leger" wrote:
>> Keith, Rik and anybody else,
> While you are paying attention to those AP photos could you scrutinise
> the beaded appearance along the right hand edge of AP?
> ...
> Setting aside camera artifacts, which are an important but separate issue,
> I am beginning to think that we too readily dismiss some things in images
> as diffraction artifacts when in fact they might help in the deduction of
> underlying structure. I posit that this prejudice has arisen.... Selwyn,
I'm specifically NOT going to tackle this issue.

As a bit of philosophy, it seems clear that almost anything we see in an image tells us *something* about the specimen. But determining exactly *what* it tells us can be a very difficult problem.

The more an image has been affected by diffraction, by non-uniform illumination, by lens aberrations, and by interaction between various parts of the specimen, the harder it becomes to work backward from seeing an image to determining what structure produced it. Even in the finest mathematical theory, with perfect understanding of the image-forming process, there are strict limits to what can be inferred.

It may be productive to think in terms of "what can we reliably infer from the images that we see?"
In many cases, I'll bet this turns the phrase "dismiss as artifact" into "ignore as unreliable information".

I don't know whether that helps much, but it's a different spin on the problem, anyway.

If you can identify specific cases or methods to reliably infer more structure from heavily diffracted images, it seems like that would be a valuable contribution. (I started to write "a *welcomed* contribution", but then I realized that was probably way too optimistic. Active resistance might be more likely!)

-Rik

Dear Selwyn,
If you are trying to get the most from you camera you need to save the image in the raw formant. You can't turn off all the image processing in a consumer camera any other way. The way I got started on looking at artifact on the Coolpix 995 and 4500 was seeing the differecne in the image processing a Coopix 950 and and 995 did not the artifacts of the 995.

The image a Coolpix 950 saves looks the same as the one you see on the live view and that's not the case for a 995 or 4500.

If you are trying to image small features you may not need all the pixels your camera has but you can't disregard the artifacts that JPG's and the camera introduce. You have to decimate the data yourself to know what is going on. If you rely on a camera designed for something else to do it for you you don't know what you have.

I haven't been able to find the spectra of the 4500 sensors sensitivity. I know using the YCM filter set you have to have a 4th image for black you don't have to have with RGB set, I don't know what would happen with YCMG filters.

Gordon

"A. S. St Leger" wrote:
> I don't use the highest camera mode but the next best which gives
> 1600x1200 (24 BPP) pixels. I tend to the view that so much of these
> photos is repetitive data that Jpeg compression looses little if any
> information. Perhaps, when pushing the camera to its limits I should
> use the highest mode with all its consequent inconvenience.

Hi Rik, Selwyn,
What we see down a microscope and call an image is an artifact. Produced by a spectrum of light, through an optical system, as you have pointed out, riddled with defects and described by mathematics that does not tell half the story! L. C. Martin in his excellent book on the 'Theory of the Microscope', points out that a comparison between classical optical theory and a full expression using eletromagnetic theory, shows that the EM theory predicts better resolution. Umm! He does not specify which EM theory, finite element or method of moments. Umm!

Exactly how long is a piece of string? The way that marvellous parallel machine we call the brain composes, from the visual information, what we call an image is even less well understood than our grasp of (largely) 18th century mathematics. There is a marvellous line from 'The hunting of the Snark', by Lewis Carroll. 'What's the use of Mercators North poles and Equators, Tropic zones and meridian lines?' Thus the Bellman would cry and the crew would reply, 'They are merely conventional signs!' The conventional signs we use to describe what we see and wish to analyse are seldom fully up to the task.

As I grow older I tend to echo the founding member of the Royal Microscopical Society, 'God Bless the Microscope!' With it we recover that child's sense of wonder at something newly discovered, and not understood. Which is why we are all here. A world in which everything is understood has no sense of wonder and is ultimately sterile. So God Bless the members of the Yahoo Microscopy group, we are part of a wonderful, "what if" tradition that keeps that fundamental curiousity alive!

Cheers, Merv

Dear Gordon,
Many thanks for the advice. I shall explore the difference raw imaging makes when pushing my optics to the limit. Most of the time, particularly when dealing with protists, I shan't be anywhere near the limit but AP is fascinating me at present as is DLC (about which I remain agnostic but not ready to dismiss.)

Regards, Selwyn

Hi Merv,
You state the sense of wonder, and expectation of the unanticipated, elegantly. Would that we keep our minds open. I suspect that we haven't heard the last word on optical microscopy despite the fact that many draw the line under Abbe. It's minds like Carol's (Dodgson's prediliction for logical conundrums) that keep us from conformity.

Incidentally, I am working my way through Martin's book now. It is very well written and seemingly thorough.

Regards, Selwyn

Dear Selwyn,

I agree with you in principle but it is real mess if you try to go there. No camera can capture all you eyes and mind can see. Depending on you vision and memory you can see more much than the few dimensions the camera can capture. You brain is a very very good adaptive filter that effortlessly fills in some blanks and sis in the data.

Likewise a camera sees things you eyes and mind over look. The telephone pole growing out of your wife's head in a picture and high lines and jet air craft trail that mar the perfect landscape you saw in you mids eye when you took the picture.

Add to that the grain of photographic film or pixelization of a monochrome camera or pixelization and the Bayer or other patten of the color filter plus what ever the software done in the dark between time the image is snapped and it it written to memory to compensates for the pixelization, Bayer pattern, alising, bumping up he contrast and what ever else the camera maker does to make the image look better so his camera sells better. And I can't sort it all out. I will make a raw image and process it from there at least I know whats been done the image that way.

best regards, Gordon

Selwyn,
If you are pushing limits your eyes are the test. Then you try to get a picture of what you saw.

When you try a technique like DLC you need something to compare it with. COL is the cheapest and easiest way to increase contrast in way that is predictable and you can find help with and we know behaves in predictable way.

To know if something is better you have to have something to compare it with or you end up as my old journalism professor would write on a paper that used better with something to compare it to he would right in blue, "Better than what? Old used Coffee grounds?"

Gordon

"A. S. St Leger" wrote:
> Many thanks for the advice. I shall explore the difference raw
> imaging makes when pushing my optics to the limit.

Hi Selwyn,
I'm not in a position to say anything about the usefullness of these effects, I'll keep in line with Rik's comments. But check out this link:
  Link : http://www.jcb.org/cgi/content/abstract/164/3/337

René.

Dear Gordon,
I take your points entirely. Certainly I see (I believe) more than I acn photograph and don't take photos until I have seen something interesting. Like others with consumer cameras I have to take multiple photos at different microscope focus because the camera cannot reliably be set up exactly enough parfocal for truly critical observations. I have another technical impediment in that with 16x oculars what one sees (re AP dots) is still barely enough magnified. Apparently my Zeiss Standard 16 cannot take an optovar in addition to the DIC intermediate tube.

You are right about comparison being a key issue. Previously I have suggested with respect to DLC that we compare what can be seen with lower NA objectives against what is expected with higher NA objectives. Comparison with oblique, COL and darkfield would be very valuable too. I am beginning to suspect, based on visual observation, that some of the ehanced resolution claimed for DLC may be seen with COL but was previously dismissed as diffraction artifact. That said an order of magnitude enhancement of resolution (taking the term literally) would mean that an objective with NA 0.3 could resolve approx 0.09 microns which seems far fetched and which incidentally would require considerable magnification for either the eye or a camera to detect. Maybe increased resolution (of whatever order) occurs mainly with immersion objectives. None of these are matters I can take further.

Regards, Selwyn

Hi René,
Thanks for that fascinating article. A quick initial skim through indicates that Moire effects are more than curiosities. For instance someone visualising dots on AP by Moire is genuinely seeing structure (albeit magnified) and before the presence of dots was known could have made useful deductions/speculations about their presence and spacing. Perhaps they did, I just haven't come across a reference.

Regards, Selwyn

Once, a few years ago, I referred to DIC, Hoffman and Phase contrast etc. as Real Time Image Processing and to a simple setup using transmitted light as 'ordinary light microscopy.'

Well, any set of lenses used to image an object is image processing, but I drew a 'Traditional' line between it and DIC etc. I've been reading a lot about these techniques in Biophotonics and elsewhere recently and come to the conclusion that since every method of imaging is processing it makes no difference at all what you use. But utilizing special configurations to enhance images and improve resolution is not 'breaking the Abbe limit' the limit still holds. The methods are quite different and the limit simply doesn't apply.

I posted a link to an article last year in which the authors attained resolutions of 50nm or less -- I can't remember the details now.

Don W

I understand that the Coolpix 'raw' format image is a basic bitmap without any data compression. But do you have 3 bytes per pixel (24- bit color, Bayer interpolated) or 1 byte corresponding to the 8-bit dynamic range for each R,G, or B photosite?
For the Coolpix 4500 at maximum resolution (2272x1704 pixels) the file sizes would be ~ 11.6MB and ~ 3.9MB respectively.

I expect it's the interpolated data, even for monochrome mode as explained by BobC in a previous post, but I would like to be sure - especially in view of the comment re turning off all the image processing.
Because if this is the case, any artifact that is introduced is due to JPEG compression as the Bayer array processing is out of our control (I think). Or to put it another way, we do not have access to the "really raw" data associated with each photosite, as there is always some per pixel data processing that takes place internal to this consumer camera. This seems to be a key difference as compared to a pro camera using software on a PC (for instance) that handles all data processing based upon the raw photosite information.

Btw, it soon becomes evident that we need FTP to transfer raw image files for further analysis. But it's convenient that the 4500 stores the raw image in the industry standard TIFF (uncompressed) format.

With respect to artifacts in general, the one place you do not need them is within the primary image data. For any data processing that follows, it's the ol' "garbage in gives garbage out" problem and it usually gets worse for each processing step. Therefore I believe that Gordon and Rik make very good points re raw format and utilizing the sensor's native resolution to acquire the primary image data.

-Keith"Gordon Couger" wrote:
> If you are trying to get the most from your camera you need
> to save the image in the raw format. You can't turn off all the
> image processing in a consumer camera any other way.

Hi Rene,
Thanks for the link, Moire fringes are quite fundamental to understanding how diffraction structures forms an image. If we consider that all parts of the image are composed of Airy ring patterns - J. W. Gordon's 'anti-point', then any displacement of one ring system with respect to another produces parallel fringes orthogonal to the direction of displacement. These parallel fringes spacing is proportional to the distance of displacement between each antipoint in the Fourier plane. This system was first analysed by Arnold Schuster at the University of Manchester, and the ideas were further developed by Raman and Datta in the 1920s.

The 4pi guys seem to have picked up on this idea and have used it very successfully to bring signals out of the diffraction 'noise'. The important idea is that the diffraction artifacts limit the resolution by spreading out the information around each point in a spatially defined and periodic way, related to the aperture of the objective and the wavelength of observation.

However, Abbe's analysis is 2D, and is therefore contaminated by confocals we cannot resolve and remove. With the two objective approach, first used on interference microscopes in the 50s and modern computer based analysis, it can be done. Hence the results seen in the articles published.

As I pointed out some time ago, a polarisation microscope can resolve in the vertical axis to lattice spacing distances of a few Angstroms. It is only the superposed diffraction pattern induced by the instrument that prevents better lateral resolution, across the aperture. Remove that, and the Abbe limit is no limit at all!

Cheers, Merv

Thanks Merv. I hardly know what you're talking about ;-) but I realise it's a long way off from real time visual microscopy, well who knows... I think the authors used algorithms designed for showing what they think it should show (ie microtubular strands). Not exactly cheating but not blind either.

So is the 4Pi possible with double objective/combining real time through the eyepieces? I mean the system with reduced axial resolution?

René.

Hi Rene,
I hardly know myself, these days....

Some Spanish guys published a series of papers using spatial filtering in real time - I will recover the links and post them. You are right that the algorithmic approach was 'tweaked' till they saw what they wanted. That is inevitable with the crudity of the algorithmic approach, you only have limited degrees of freedom you can programme - not much different to Royal Rife in practice.....

Best regards, Merv

Hi Merv and Rene,
Yet another very interesting post to chew on! Thanks.

> Thanks [Rene] for the link, Moire fringes are quite fundamental
> to understanding how diffraction structures forms an image.

With very limited hobby time right I can't keep up with the info flow. But I seem to remember some light microscopy images showing very clearly a diatom valve's pore structure that relied upon "Moire magnification". The effect was akin to having a "lens" attached to the diatom valve to give a real, effective magnification of about 2000x.

I could dig this info out again and prepare a webpage, if the group is interested.

> ... As I pointed out some time ago, a polarisation microscope
> can resolve in the vertical axis to lattice spacing distances
> of a few Angstroms. It is only the superposed diffraction
> pattern induced by the instrument that prevents better
> lateral resolution, across the aperture.

This brings to mind an article that was referenced during one of the DLC discusssions, now archived at:
  Link : http://www.kscitech.com/MicGroup/Msg_07
A group at one of the Max Planck Institutes demonstrated 30-35nm resolution for some features of a bacteria preparation. It's based upon flourescence microscopy, but uses a vertical-lateral scanning technique with the highest resolution being in the vertical axis (optical axis).

Best regards, Keith

Hi Keith,
Thanks for the link, the resolution of the cell wall using fluorescence and Helmhotz conditions is right at the limit of what can be achieved. This is leading edge stuff as it is vital to understand how the complex chemistry of the cell wall functions in live cells. It is sobering to think that what happens across a 30nm thick lipoprotein layer in terms of electrochemistry makes the difference between a healthy or sick cell. This understanding is the key to successful chemotherapy in terms of cancer, and gene therapy. Hence the burgeoning interest in 4pi techniques. It is important to make clear that no laws of physics are being broken, the Abbe limit is based on a particular set of mathematical conditions, tailored to fit a set of instrumental circumstances. Yes, resolution in a 2D sense is diffraction limited, but it can be instrumentally removed if you have 3D information; if you have all the phase and amplitude information in diffraction for a given spherical surface surrounding the object, resolution is infinite, hence the name 4pi. Well, in practice it is not quite as easy as that, but it is a nice idea!

Cheers, Merv

Dear Keith,
Excuse me if I am repetitive, and probably boring, but given the fact that the image taken by the camera includes all the distortion from all the optical path from the lamp to the camera lens, the only possible comparison is between two complete systems, and for this purpose I do not see any better way than to take a picture of a blade edge, which, under ideal conditions should show a dark and a clear field, with very few pixels involved in the transition. Fixing other parameters, the less the pixels the better the system. I have already some data from my tests and those of some italian friends. May be I am wrong, but a really quantitative evaluation is possible by this way.

Salvatore

"scitech200" wrote:
> With respect to artifacts in general, the one place you do not need
> them is within the primary image data. For any data processing that
> follows, it's the ol' "garbage in gives garbage out" problem and
> usually gets worse for each processing step. Therefore I believe
> at Gordon and Rik make very good points re raw format and utilizing the
> sensor's native resolution to acquire the primary image data.

Dear Salvatore,
I think I understand your point with respect to a well defined reference blade edge. John Hart brought this basic issue up with the group some time (maybe years!) back, and there were some test results presented.

It seems that for a consumer digital camera there is an 'anti- aliasing filter' involved which removes high frequency components (in a Fourier domain) as this processing gives, in general, a "more pleasing" picture. A light microscopist, however, may consider it as a blurring of the image.

I remember taking some test photos of a lithographic mask (high resolution B&W film) and the very sharp transition between 'black' (opaque) and 'white' (transparent) background for the film were transformed into a linear array of pixels having shades of gray, as well as the black and white pixels. As I recall, this array was at least 4 pixels in length for a SONY F505 2.1 mpixel camera.

But as the pros using high resolution DSLRs have very clearly demonstrated, it doesn't matter if such an array spans pixels in an image that contains significantly more pixels than necessary to define a 'sharp' feature. I'm not sure that it's the correct term, but I know that this is often referred to as over-sampling. RikL mentioned this in a recent post when he suggested a rule of thumb for determining how many pixels are "safe" to obtain an accurate image.

Now when we resample to reduce the size of an image for presentation on the web, etc....

I'm sure that other members can add to this discussion.

Regards, Keith

"Salvatore Previtera" wrote:
> ....comparison is between two complete systems, and for this
> purpose I do not see any better way than to take a picture of
> a blade edge, which, under ideal conditions should show a dark
> and a clear field, with very few pixels involved in the transition.

"scitech200" wrote:
> I understand that the Coolpix 'raw' format image is a basic bitmap
> without any data compression. But do you have 3 bytes per pixel (24-
> bit color, Bayer interpolated) or 1 byte corresponding to the 8-bit
> dynamic range for each R,G, or B photosite?

Raw data from a camera is one number per pixel, before Bayer interpolation. It is also generally wider than 8 bits - usually 12. It may be losslessly compressed. The compression means that the number of bytes of data varies with image content. I don't know what the Nikon 4500 specifically writes, but this is generally true of RAW output from many manufacturers.

> I expect it's the interpolated data, even for monochrome mode as
> explained by BobC in a previous post, but I would like to be sure -
> especially in view of the comment re turning off all the image processing.

No, the whole point of RAW is to avoid almost all the image processing done by the camera, so you can do it on your computer later with different parameter settings. This includes the Bayer demosaicing as well as sharpening, converting 12 bits linear to 8 bits non-linear, and lossy compression - none of these have been done to raw data.

> Or to put it another way, we do not have access
> to the "really raw" data associated with each photosite, as there is
> always some per pixel data processing that takes place internal to
> this consumer camera.

You don't have control over the analog amplifier gain ahead of the A/D converter, but essentially all other processing is done on your computer. There are several vendors that offer commercial raw processing software, plus you can write your own if you want (look up the "dcraw" program as a starting point).

> Btw, it soon becomes evident that we need FTP to transfer raw image
> files for further analysis. But it's convenient that the 4500 stores
> the raw image in the industry standard TIFF (uncompressed) format.

TIFF is yet another output format supplied by some cameras, but it's not RAW. TIFF output has had all of the demosaicing, sharpening, tone curve shaping, and 12-to-8 bit conversion done, just like a JPEG. The only thing TIFF output avoids is the lossy JPEG compression. There's much more difference between RAW and TIFF than between TIFF and JPEG.

Dave

"dmmartindale" wrote:
> It is also generally wider than 8 bits - usually 12.
> It may be losslessly compressed.

Nikon says that they compress from 12 bits down to 10 in the raw image.

> TIFF output has had all of the demosaicing, sharpening, tone
> curve shaping, and 12-to-8 bit conversion done, ...

Close, but TIFF does not necessarily compress to 8 bit. TIFF can support an arbitrary number of bits per pixel and often is set to 16 bit data, or 48 bits per pixel.

The degradation resulting from JPEG compression is determined by the image content. A continuous tone image with no abrupt brightness changes (and microscope images are often "soft" like this) can tolerate a much higher degree of compression than an image with abrupt brightness/color changes. The worst performance for JPEG would be with line art.

Dave,
You wrote:
> TIFF is yet another output format supplied by some cameras, but
> it's not RAW. TIFF output has had all of the demosaicing,
> sharpening, tone curve shaping, and 12-to-8 bit conversion done,
> just like a JPEG. The only thing TIFF output avoids is the lossy
> JPEG compression. There's much more difference between RAW and
> TIFF than between TIFF and JPEG.

Thanks very much Dave for this explanation. Some members may consider it to be nit-picking, but I believe it's very important that we all understand details such as these during a group discussion - for instance as related to the Coolpix 4500 camera.

Regards, Keith