A.pellucida - Blue LED

Something else to consider when thinking about illumination for photography is the sensor in the camera. A couple of issues come to mind:
CCD sensors typically have their peak sensitivity around the yellow/green area, falling off rapidly as wavelength decreases (blue) and less rapidly towards the red/infrared wavelengths. For an example curve for a monochrome CCD:
  Link : http://products.sel.sony.com/semi/PDF/ICX285AL.pdf (page 8)

This sensitivity is also modified by the colour Bayer matrix used in most colour cameras, resulting in curves as shown in:
  Link : http://products.sel.sony.com/semi/PDF/ICX285AQ.pdf (page 8 again)

The reduced sensitivity at shorter wavelengths means that a higher intensity light source will be needed in order to prevent noise resulting from the camera needing to amplify the signal.

    R G R G R G R G
    G B G B G B G B
    R G R G R G R G
    G B G B G B G B

There are twice as many green pixels as there are red or blue, so the camera's resolution in green will be twice that in the other colours.

    C Y C Y  C Y C Y
    M G M G M G M G
    C Y C Y  C Y C Y
    G M G M G M G M

But even with these filters there are more sites that transmit green than there are red or blue.

I don't know how these facts actually translate in practice to the maximum resolution achievable in a digital photomicrograph - maybe it's just a question of using a short wavelength, turning up the brightness and adjusting the image scale appropriately to match the Bayer filter matrix - but I think it's something that needs to be borne in mind when illuminating with 'pure' colours.

Cheers, Jon

Hi Selwyn,
Most, if not all, consumer digital color cameras use the RGB Bayer filter, as described by Jon.
This is implemented by placing red, green and blue filters over the photo-sensor (CMOS or CCD) sites - as briefly described in a webcam review:
  Link : www.kscitech.com/Webcam

The actual RGB color information transferred from the camera corresponding to each photo-sensor site is actually computed using the R,G and B values of neighboring sites. We should note that the number of photo-sites (pixels) is defined by the camera's resolution. Thus a typical 24-bit RGB color for a pixel from the camera is an approximation to the 'true' color at any pixel location within the camera's sensor array. It can take some time to grasp the full significance of the Bayer filter and the digital processing (commonly referred to as de-mosaicing) that takes place inside a consumer camera!

Now with research grade color cameras it's a different story. Some have 3 CCDs to pick up the 'true' R, G and B components for a pixel. Others use a rotating filter wheel with a single CCD (monochrome camera) to obtain the same component information.

I don't know how to describe the Coolpix B&W mode, but the Bayer filter is still involved and so we can only assume that a different algorithm is used to compute the so-called B&W colors.

Now I have to go back and check the camera that Peter used for his fine A.pellucida image - it may have been a monochrome unit. If so, he's avoiding any possible loss of resolution due to the approximations involved with the Bayer filter computations.

I seem to remember Frez presenting some great monochrome images, way back, but other members may have more on this subject.

Regards, Keith

"selwynstleger" wrote:
> I use a Coolpix 4500....
> Finally, there is a B&W mode such that the images cannot be divided
> into RGB....

Kind regards, Marien

"Jon Grove" wrote:
> Something else to consider when thinking about illumination for
> photography is the sensor in the camera.
....
> A different issue is the Bayer matrix itself. Many cameras
> have a matrix structure like:
> RGRGRGRG
> GBGBGBGB
> RGRGRGRG
> GBGBGBGB
> There are twice as many green pixels as there are red or blue,
> so the camera's resolution in green will be twice that in the other colours.

Note also, in some cameras *all* of the sensor sites are sensitive to green, with the "red" and "blue" sites just relatively more sensitive to those colors. See:
  Link : http://tinyurl.com/cb8m4

> ...maybe it's just a question of using a short wavelength,
> turning up the brightness and adjusting the image scale
> appropriately to match the Bayer filter matrix...

Good summary! Be conservative, treat your 3K x 2K color camera as if it shoots a 1.5K x 1K blue image, and you should get no surprises.

Imaging a 150 micron field at this image size would give 0.1 micron per pixel, not really enough to assure good sampling at 0.2 micron resolution.

But zooming the camera lens a bit, say 2X, should guarantee that you're seeing all there is to see.

-Rik

Hello Keith,
With all the color cameras I have had not really good pictures. I tried it with:

Canon EOS 10D
Olympus C3040
Olympus C2020
WebCam Philips 680K
WebCam Philips 840

It does not work. The Bayer pattern and the next processor in your camera makes it nearly impossible.

After I take a monochrome camera with a Sony Chip ICX424AL, I have had the success.
The camera is from Lumenera. The spectral sensitivity for 400nm is near 60% !

Think about.

Hello Peter,
Thank you once again for your input.

When I first met the Bayer filter I got all messed up trying to understand the basics, and I'm still not sure that I have everything sorted out!

Photomicrography is a complex situation, with many variables. But what you have demonstrated with the monochrome, relatively low resolution camera for this specialized imaging application makes a lot of sense.

Then Jon's comment: "...adjusting the image scale appropriately to match the Bayer filter matrix" brings up yet another issue - matching the camera sensor's pixel size to the smallest resolvable feature. I still struggle with the "more pixels is not necessarily better argument". It just seems that over-sampling with a high resolution color camera should minimize loss of resolution, and the introduction of artifacts, due to Bayer filter processing. But....

Regards, Keith

"Peter" wrote:
> It does not work. The Bayer pattern and the next processor in your
> camera makes it nearly impossible.
> After I take a monochrome camera with a Sony Chip ICX 424 AL ,
> I have had the success....
> Think about.

Hi Rik,

Thanks for the link to the Canon colour sensitivity test - I hadn't seen that one.

But I wonder if your conclusion that 'all sensor sites are sensitive to green' can really be drawn from this experiment. I would say that the sensor sites have filters whose spectral response appreciably overlaps that of the Wratten #61 filter used in the test, but the overlap is most likely in the yellow and bluegreen (for red and blue sites respectively) wavelengths.

So the resolution achievable will depend to an extent on the overlap between the spectrum of the light source (LED) and that of the sensors in the camera.

Some other things that come to mind:
- In an RGB Bayer array it is not uncommon for the two different 'G' sites to use different filter types - which complicates things further!
- To help with the problems of undersampling the colour cameras often have a 'low-pass filter' in front of the CCD which effectively blurs the image by a small amount, so that details with dimensions comparable to a single sensor site do not introduce colour artifacts.
- I think for ultimate resolution you need to use a monochrome CCD.

But since a colour sensor is generally the only practical/affordable option then it would pay to perform some experiments with different illumination colours and image scales. I'll be very interested in the results!

Cheers, Jon.

Hi Selwyn,
I think your assessment of Black and White image made with a camera with a color camera with a Bayer filter is right. I would compare the image I got with ImageJ with what the 4500 made as a monochrome image. There should be a better way to get a gray scale image than just splitting RGB or CYM.

If using narrow band light one should use a monochrome camera so there are no problem with the red, green and blue pixels and problems inherent with the filters over the various pixels. A monochrome camera with no Bayer filter to reduce the resolution and with the pixels laid out in rows and columns works much better for monochromatic light. Up to 3/4 of the [color] camera's pixels may be useless using narrow band monochromatic light.

No filters are needed anywhere in the system for narrow band monochromatic light. The Osram document:
  Link : http://catalog.osram-os.com/media/_en/Graphics/00039053_0.pdf
shows the spectral sensitivity of the eye and some sensors.

There are a number of things that make up the ability to resolve detail, see and photograph it.. The resolving power of the microscope, the contrast of the image and the ability of the recording device to see the resulting image. One can't be maximized without loss in one of the other areas. Table 2 in Nikon's Microscope U's page:
  Link : http://www.microscopyu.com/articles/formulas/formulasresolution.html
shows that resolution increases directly as the wavelength shortens so as with everthing else the same 360nm light resolves feature nearly half the size 700 nm light. While it is not in the table the value for the size 350 nm light would resolve would be exactly half the size of 700 nm light. So we see why we are all taken by blue and violet light.

However, all things are never equal. Without question our eye is the best device for seeing and interpreting detail. It works best with green light and the microscopes we have are almost without exception optimized for green light. So when we use tools that are optimized for the convenient 546 nm green light a mercury arc emits. There was no other source of powerful green light economically available until fairly recently. Using 340 nm light we should be able to resolve a .18 nm feature incited of a .29 one. But we sure don't seem to be able to realize it in practice on a diatom and consumer camera.

Our eyes, microscope optics and most of the cameras we use don't preform as well in the shorter wave lengths as they do at the green mercury line most optical houses used for light that designed microscopes. In the ongoing saga trying to resolve features of A.pellucida, an endeavor that has driven microscopists for over 100 years to perfect their technique a combination of contrast, resolution and the ability to capture what is seen in the eyepieces challenges every one that tries it to their very limit.

Contrast is one of the first things to suffer when we get outside the best performance of the scope. I think all the tricks with light to increase contrast come at some cost in resolution.

No camera can capture what we can see with our eyes and brain. Not only can the eye see more than the camera the mind can assemble a 3D image and integrate information over time and many tries of different lighting in an hand drawn image. To my mind Leeuwenhoek's drawings prove that eye is better than a camera. Today even the one that draws the image can't be sure how much he really saw and how much he drew from previous knowledge. In Leeuwenhoek's case there was no previous knowledge to draw on and later observations agree rather well with his drawing.

I am working with lighting right now and have 2 monochrome Video cameras that need to be tested. I can test them against a CoolPix 995 and to see how a Bayer filter compares with a camera without on the same image. I have narrow band green and violet filters. The difference in pixel count is not as bad as it seems as it is hard to find a low enough powered eye piece to fit the image on the 1/2 or 1/3 inch CCD of the video camera. It is hard to spread the image on the Video camera over as as few pixels as it is on the CoolPix.

I plan to start at the light source and go to the scope. Ted Clarke's work in:
  Link : http://www.microscopy-uk.org.uk/mag/artsep01/tcmacro.html
looks like a good way to go to me. I think I will try a 25mm cine lens first to see if a smaller lens will work but I can go up as big as I need to.

I have tried starting at the camera before. I am going to do it differently this time. Starting with lighting and going right up to the camera getting it all right as I go.

Gordon

"selwynstleger" wrote:
> I use a Coolpix 4500. In colour mode one can split the images, in,
for instance, ImageJ, into RGB. Would I be correct in thinking that
> the blue image is highest resolution (with say apo or flourite
> objective) than the others even though a broad blue spectrum is covered?
>
> If one uses a blue filter then presumably the red and green images
> should be discarded too, would you agree?
>
> Finally, there is a B&W mode such that the images cannot be divided
> into RGB. Presumably using this with a blue filter is no different
> from using the blue image from RGB mode?

"Jon Grove" wrote:
> ...So the resolution achievable will depend to an extent
> on the overlap between the spectrum of the light source
> (LED) and that of the sensors in the camera.

I agree completely. My simple statement essentially defines "green" to be "whatever gets through a Wratten #61 filter given a continuous spectrum source". It's quite possible that there is no single wavelength that all the sensors can see, and there may be many color bands that can be seen by only one.

--Rik