Loading ...
Sorry, an error occurred while loading the content.
 

Re: [ATM] Lerch's Robo Foucault VS. Interferometry, revisited

Expand Messages
  • Michael Peck
    ... James: A couple quick suggestions: Doesn t your Foucault analysis algorithm look at a fairly narrow band of pixels along the horizontal axis? If so the
    Message 1 of 20 , Feb 1, 2005
      At 14:31 1/31/05, James Lerch wrote:

      >If you'd be so kind as to have a look, and let me know what you think, I'd
      >appreciate it.

      James:

      A couple quick suggestions:

      Doesn't your Foucault analysis algorithm look at a fairly narrow band of
      pixels along the horizontal axis? If so the comparison you want to make is
      to a cross section along the same diameter (first graph), rather than one
      constructed from radial Zernikes only.

      Why not reflect the surface error line from Foucault onto the left side of
      the cross section graph so you can compare the whole diameter. Simulations
      that Jim Burrows and I have done suggest that if the mirror isn't
      symmetrical (and they never are exactly) Foucault results should more or
      less split the difference between the left and right halves. That appears
      to be pretty much the case in most of your examples.

      I've been thinking a little bit about how to give summary quantitative
      measures of how good the correspondence is between Foucault and
      interferometry, and so far have the following ideas:

      a) The (area weighted) rms difference between surface error profiles.

      b) The rate of "false positives" and "false negatives". Let's assume for
      now that interferometry is perfectly accurate (we know it isn't, but never
      mind that). If Foucault says a mirror is good enough by some criterion and
      Interferometry says it isn't call that a "false positive." On the other
      hand if Foucault says it's not good enough by the same criterion and
      Interferometry says it is call that a "false negative." The sum of those
      two is the total error rate.

      Overall it looks as though you're getting more consistent results these
      days. Good work.

      Mike Peck


      ------
      Michael Peck
      mpeck1@...
    • atmpob
      James is using my openFringe software to create those profiles. It does not reflect the robo profile so he has no way to make it do so. However openFinge does
      Message 2 of 20 , Feb 1, 2005
        James is using my openFringe software to create those profiles. It
        does not reflect the robo profile so he has no way to make it do so.

        However openFinge does have a display I just told him about. It can
        average the left and right profiles from the Zernikes and display
        that as one of the profiles. I think you and/or Jim Burrows thought
        that perhaps that is what Foucault sees. He can turn that on and
        recreate the graphs. You will see that in general robo matches the
        symetrical terms only better than it matches the left and right
        Zernike average. I think that will show you what you requested but
        in a slight different format.

        It could be that we still are not using enough Zernike terms to get a
        good match. Both James and I have noticed that Zernike version of
        the center of mirror B does not match what we see in Foucault
        images. In those we see a sphere at the center. But the Zernike fit
        show the center to be too low. A sphere should be too high shouldn't
        it?

        I found an interesting thing working with Mirror B. Roger C. shows
        an average profile of all the diameters in his round robin report.
        That matches very close to what both James and I have measured using
        only the spherical terms. He doesn't say how he created it but I'm
        guessing he used only the spherical terms. When I turn on all the
        terms I find that the robo profile matches the 45 deg diameter better
        than either the horizontal or the vertical diameter of the igram.
        That tells me that the robo is some sort of average between both the
        vertical and horizontal diameters.

        Dale Eason

        --- In atm_free@yahoogroups.com, Michael Peck <mpeck1@i...> wrote:
        > At 14:31 1/31/05, James Lerch wrote:
        >
        > >If you'd be so kind as to have a look, and let me know what you
        think, I'd
        > >appreciate it.
        >
        > James:
        >
        > A couple quick suggestions:
        >
        > Doesn't your Foucault analysis algorithm look at a fairly narrow
        band of
        > pixels along the horizontal axis? If so the comparison you want to
        make is
        > to a cross section along the same diameter (first graph), rather
        than one
        > constructed from radial Zernikes only.
        >
        > Why not reflect the surface error line from Foucault onto the left
        side of
        > the cross section graph so you can compare the whole diameter.
        Simulations
        > that Jim Burrows and I have done suggest that if the mirror isn't
        > symmetrical (and they never are exactly) Foucault results should
        more or
        > less split the difference between the left and right halves. That
        appears
        > to be pretty much the case in most of your examples.
        >
        > I've been thinking a little bit about how to give summary
        quantitative
        > measures of how good the correspondence is between Foucault and
        > interferometry, and so far have the following ideas:
        >
        > a) The (area weighted) rms difference between surface error
        profiles.
        >
        > b) The rate of "false positives" and "false negatives". Let's
        assume for
        > now that interferometry is perfectly accurate (we know it isn't,
        but never
        > mind that). If Foucault says a mirror is good enough by some
        criterion and
        > Interferometry says it isn't call that a "false positive." On the
        other
        > hand if Foucault says it's not good enough by the same criterion
        and
        > Interferometry says it is call that a "false negative." The sum of
        those
        > two is the total error rate.
        >
        > Overall it looks as though you're getting more consistent results
        these
        > days. Good work.
        >
        > Mike Peck
        >
        >
        > ------
        > Michael Peck
        > mpeck1@i...
      • atmpob
        For anyone who wants to analyze James data or do similar expriments at other diameters I can send you openFringe. That is what he used. You can use it to do
        Message 3 of 20 , Feb 1, 2005
          For anyone who wants to analyze James data or do similar expriments
          at other diameters I can send you openFringe. That is what he used.
          You can use it to do other analysis using the Zernike files, Fringe
          files and Igrams he has an his site.

          With it you can see what other profiles look like. Turn on and off
          various Zernike terms to compare the results. You can show simulated
          Foucault grams from the data. You can turn on flip and diff zone
          measurements to show where a null would be and then change Zernike
          terms to see how that effects the null position. You can create star
          test images. You can import sixtests surface profiles to compare.

          You can do a lot of neat stuff. It is in beta so sometime it does
          not work right. But James is getting good results from what he tries.
          Star test images have not been verified but they look cool.

          Runs on Windows 98E and above. Oh, it can also analyze interferograms
          which is what it was intended for.

          Dale Eason
        • Michael Peck
          ... can ... thought ... Not an average necessarily. That s why I used the intentionally imprecise phrase more or less splits the difference. ... the ... The
          Message 4 of 20 , Feb 1, 2005
            --- In atm_free@yahoogroups.com, "atmpob" <atmpob@y...> wrote:
            >
            > > However openFinge does have a display I just told him about. It
            can
            > average the left and right profiles from the Zernikes and display
            > that as one of the profiles. I think you and/or Jim Burrows
            thought
            > that perhaps that is what Foucault sees.

            Not an average necessarily. That's why I used the intentionally
            imprecise phrase "more or less splits the difference."

            > That tells me that the robo is some sort of average between both
            the
            > vertical and horizontal diameters.
            >
            >

            The intensity distribution in the shadowgram along a chord
            perpendicular to the knife edge only depends on the wavefront error
            along the corresponding chord on the mirror. If what goes into your
            foucault measurements is data from a narrow horizontal band of pixels
            you are measuring a surface cross-section along that horizontal band.
            At least if you're measuring anything meaningful at all.

            In my algorithm I used a wedge of pixels within +- 45° of the
            horizontal axis. That was intended to provide a mean radial profile,
            which it seems to have done in the one case where I have both
            Foucault data and interferometry.

            This seems like another reason to produce a simple quantitative
            summary measure of how close two curves are. I don't trust my
            eyeballs to make that judgement.

            Mike Peck
          • James Lerch
            ... From: Michael Peck ... Indeed it does, 11 pixels where each Couder mask hole should be. The 11 pixels are arranged in an arc
            Message 5 of 20 , Feb 1, 2005
              ----- Original Message -----
              From: "Michael Peck" <mpeck1@...>

              > James:
              >
              > A couple quick suggestions:
              >
              > Doesn't your Foucault analysis algorithm look at a fairly narrow band of
              > pixels along the horizontal axis?

              Indeed it does, 11 pixels where each 'Couder mask' hole should be. The 11
              pixels are arranged in an arc following the zone radius, and centered along the
              horizontal diameter.

              >If so the comparison you want to make is
              > to a cross section along the same diameter (first graph), rather than one
              > constructed from radial Zernikes only.

              Agreed, but I included both plots as I thought I heard / read somewhere that
              Foucault tests only reveal spherical Zernike terms. Keep in mind that
              Interferometery and Zernike Terms are just "Tools" to me, similar to my Miller
              Tig Welder. While I don't know 'Exactly' how my Tig welder works, I know what I
              need to do to make it weld Aluminum Vs. Steel, etc.. I have about the same
              understanding on how to go from Fringe image tracings via least squares fit to
              Zernike Terms as what happens between the 220v outlet and the Tungsten elctrode
              on my Tig welder... :)

              >
              > Why not reflect the surface error line from Foucault onto the left side of
              > the cross section graph so you can compare the whole diameter.

              Dale's software wouldn't do that, and at the time Dale's software wouldn't
              export a profile for importation into a spread sheet. Dale has since fixed the
              export problem, but I've had other fish to fry (like getting the Bino powder
              coated and ready for the WSP next week!)

              > I've been thinking a little bit about how to give summary quantitative
              > measures of how good the correspondence is between Foucault and
              > interferometry, and so far have the following ideas:
              >
              > a) The (area weighted) rms difference between surface error profiles.

              I think I know what you mean, but I don't know how to implement it... Let me
              see if I can get all 13 optical profiles exported into a spread sheet, perhaps
              that might be a start..

              > b) The rate of "false positives" and "false negatives". Let's assume for
              > now that interferometry is perfectly accurate (we know it isn't, but never
              > mind that). If Foucault says a mirror is good enough by some criterion and
              > Interferometry says it isn't call that a "false positive." On the other
              > hand if Foucault says it's not good enough by the same criterion and
              > Interferometry says it is call that a "false negative." The sum of those
              > two is the total error rate.

              Perhaps we could use the good old 0.8 Strehl (or < 20nm rms surface) criterion?

              > Overall it looks as though you're getting more consistent results these
              > days. Good work.

              Thank you, the thing that gives me the most pleasure is the fact that same Robo
              hardware and software that started all this fuss, is the same Robo hardware and
              software that's still being used to date!

              Take Care,
              James Lerch
              http://lerch.no-ip.com/atm (My telescope construction,testing, and coating site)

              "Anything that can happen, will happen" -Stephen Pollock from:
              "Particle Physics for Non-Physicists: A Tour of the Microcosmos"

              " Press on: nothing in the world can take the place of perseverance.
              Talent will not; nothing is more common than unsuccessful men with talent.
              Genius will not; unrewarded genius is almost a proverb.
              Education will not; the world is full of educated derelicts.
              Persistence and determination alone are omnipotent. "
              Calvin Coolidge
            • Michael Peck
              ... James: OK, why not? Just reading off the reported Strehl ratios from OpenFringe and FigureXP for those 13 mirrors, 8 were rated pass by both interferometry
              Message 6 of 20 , Feb 2, 2005
                At 23:39 2/1/05, James Lerch wrote:
                >----- Original Message ----- From: "Michael Peck" <mpeck1@...>
                >
                >>b) The rate of "false positives" and "false negatives". Let's assume for
                >
                >Perhaps we could use the good old 0.8 Strehl (or < 20nm rms surface)
                >criterion?
                >

                James:

                OK, why not? Just reading off the reported Strehl ratios from OpenFringe
                and FigureXP for those 13 mirrors, 8 were rated pass by both interferometry
                and foucault, 3 were rated fail by both, and there were 2 where foucault
                said pass and interferometry fail. Of those last 2 one barely failed by
                interferometry (Strehl = 0.795) but was rated quite good by foucault (0.93
                Strehl); the other one foucault said was marginally passing (0.81 Strehl)
                while interferometry calls it marginally failing (0.72 Strehl).

                If you throw my one example into the pot foucault rated it good enough
                (0.87 Strehl IIRC) and interferometry said not (Strehl 0.7).

                Mike Peck

                ------
                Michael Peck
                mpeck1@...
              • Ken Hunter
                ... DAle, Thanks for the offer to send the software... is it possible that you could send/post it to the Group? I do not know the package file size or how much
                Message 7 of 20 , Feb 2, 2005
                  --- In atm_free@yahoogroups.com, "atmpob" <atmpob@y...> wrote:
                  >
                  > For anyone who wants to analyze James data or do similar expriments
                  > at other diameters I can send you openFringe.

                  DAle,

                  Thanks for the offer to send the software... is it possible that you
                  could send/post it to the Group? I do not know the package file size
                  or how much room we have on the site but if it'll fit...

                  If not, I'll like to take a look at it!

                  Ken Hunter

                  kb7h at cableone dot net
                • jaf4oct
                  RoBo testing is a nice addition to the ATMs arsenal of tools. Many years ago I made what now I guess would be called a Robo-Lyot tester. It was very
                  Message 8 of 20 , Feb 3, 2005
                    "RoBo" testing is a nice addition to the ATMs arsenal of tools. Many
                    years ago I made what now I guess would be called a Robo-Lyot tester.
                    It was very revealing.

                    RIT did a study in the '80s comparing various interferometers and the
                    Foucault test. The end results found that a Foucault test was accurate
                    to about 1-2% of the total correction. A Robo-tester may be able to
                    improve on this figure somewhat but the Foucault test accuracy will
                    always diminish as the optic's conic constant increases. This is the
                    primary reason that compensators are used in the production of large
                    imaging systems.

                    One thing to keep in mind when using web cameras is that they all
                    produce compressed live video. That is both color bit depth and
                    dynamic range are reduced to accommodate the bandwidth limitations of
                    USB 1.1. Even USB 2.0 cameras may compress live video to reduce the
                    bus load on the PC. This will cause binning and subsequent errors in
                    the Robo-tester data.


                    Happy bit capturing

                    JAF
                  • Bob May
                    The compression on the digitized image through the USB port may match what is done elsewhere and that is to reduce the colors to 15 bits of data pwer pixel.
                    Message 9 of 20 , Feb 3, 2005
                      The compression on the digitized image through the USB port may match what
                      is done elsewhere and that is to reduce the colors to 15 bits of data pwer
                      pixel. This will show up when a subtle shade of some color is shown and you
                      will see contour lines on the image.
                      I'll note that compression like JPG won't really affect the image quality
                      (there will probably be only about a dozen pixels in the whole image that
                      have been modified by one or two counts) which won't affect the quality of
                      the image for our purposes. I'll note that JPG assumes 24 bit color.
                      Considering that we assume that the Foucault test can get down to about
                      1/20th of a wave or so on the wavefront for an accuracy limit, the RIT study
                      is about right for the quality of the test.
                      Bob May
                      http://nav.to/bobmay
                      bobmay@...
                      NEW! http://bobmay.astronomy.net
                    • jaf4oct
                      Bob, The compression loss affects the images more than you assume since the Foucault test is an intensity based test. Of the 15 bits you speak of only 5 are
                      Message 10 of 20 , Feb 4, 2005
                        Bob,

                        The compression loss affects the images more than you assume since the
                        Foucault test is an intensity based test. Of the 15 bits you speak of
                        only 5 are intensity. This allows for only 32 intensity levels. Since
                        most Robo testing does not use the full intensity range of the camera
                        and most web cams inherently have one or two bits of noise you may
                        only in reality have 3 or 4 significant bits of intensity information.
                        As a result the binning I spoke of before can have a significant
                        impact on test results.

                        Since the accuracy of the Foucault tester is dependent on the conic
                        constant of the mirror under test one cannot say in general terms the
                        Foucault test is accurate to 1/20th wave in all cases. The Foucault
                        test used on a 6 inch f/8 should, with care, be able to detect errors
                        of 1/100th wave even with less than an optimal setup according to the
                        RIT study. The same Foucault test applied to the Hubble space
                        telescope mirror would only detect errors to about 3 waves. Quite a
                        change of test sensitivity due to the large conic constant of the
                        Hubble primary. A lot of ATMs wondered why the guy who made the space
                        telescope mirror just didn't Foucault test the primary as a double
                        check. This idea did cause some humor at a SPIE meeting. With a
                        sensitivity of 3 waves even the space telescopes error would have been
                        invisible with the Foucault test. This example again shows why
                        compensators are use for those large imaging systems.

                        JAF


                        --- In atm_free@yahoogroups.com, "Bob May" <bobmay@n...> wrote:
                        > The compression on the digitized image through the USB port may
                        match what
                        > is done elsewhere and that is to reduce the colors to 15 bits of
                        data pwer
                        > pixel. This will show up when a subtle shade of some color is shown
                        and you
                        > will see contour lines on the image.
                        > I'll note that compression like JPG won't really affect the image
                        quality
                        > (there will probably be only about a dozen pixels in the whole image
                        that
                        > have been modified by one or two counts) which won't affect the
                        quality of
                        > the image for our purposes. I'll note that JPG assumes 24 bit color.
                        > Considering that we assume that the Foucault test can get down to about
                        > 1/20th of a wave or so on the wavefront for an accuracy limit, the
                        RIT study
                        > is about right for the quality of the test.
                        > Bob May
                        > http://nav.to/bobmay
                        > bobmay@n...
                        > NEW! http://bobmay.astronomy.net
                      • atmpob
                        Can you explain a little more what you mean by that? Dale Eason ... camera
                        Message 11 of 20 , Feb 4, 2005
                          Can you explain a little more what you mean by that?

                          Dale Eason

                          --- In atm_free@yahoogroups.com, "jaf4oct" <jaf4oct@y...> wrote:
                          >
                          > Since
                          > most Robo testing does not use the full intensity range of the
                          camera
                        • starzkey
                          ... the ... accurate ... By any chance, is there a published, written summary of this study that you could point me to? Having recently completed a Foucault-
                          Message 12 of 20 , Feb 4, 2005
                            >
                            > RIT did a study in the '80s comparing various interferometers and
                            the
                            > Foucault test. The end results found that a Foucault test was
                            accurate
                            > to about 1-2% of the total correction.

                            By any chance, is there a published, written summary of this study
                            that you could point me to? Having recently completed a Foucault-
                            interferometry comparo on several different mirrors ranging in
                            asphericity from an 11" F3.6 to a 10" F5, I would be interested to
                            read other's analysis and 7 description of the similarities and
                            differences seen between these two methods.

                            Scott Milligan
                          • jaf4oct
                            Dale, Webcams were designed to produce acceptable video over a large range of applications. For most users, scenes with harsh contrast (like mirror testing)
                            Message 13 of 20 , Feb 5, 2005
                              Dale,


                              Webcams were designed to produce acceptable video over a large range
                              of applications. For most users, scenes with harsh contrast (like
                              mirror testing) were to be avoided. To achieve this many webcams will
                              have a white clip and black clip. Clipping will reduce the full range
                              of the camera. The cameras DSP usually controls the clipping level.
                              You may think you have some control over this by the camera controls
                              provided by the camera manufacturer. These controls only affect the
                              DSP output not how the DSP will interpret incoming data. Camera
                              manufacturers have done a great job in finding ways to have cameras
                              mimic the human eye since that is what most users expect from their
                              cameras. Unfortunately the human eye is a poor photometer. Besides
                              clipping the video AGC and the camera gamma may not actually be under
                              your direct control as you may be only setting a range limitation.
                              Since CCDs are analog devices you are also affected by the A/D
                              converters, which may have their own logic adding unknown varibles.
                              Any of these effects can add non-linearity in the way the camera will
                              perceive the knife-edge.

                              There are so many webcams that reverse engineering them is not a
                              practical endeavor. The only real choice remaining if you want to
                              understand your particular camera limitations is to characterize it. I
                              would suggest getting a Kodak Q-60 test chart. The Q-60 has a nice
                              linear step grayscale from white to black. Don't touch the chart with
                              your fingers since body oils can change the gray step relationship.
                              You may be surprised on just how poorly your cameras live video
                              functions on this gray step chart over varying light intensity levels.
                              It will however allow you to optimize your setup to make the most of
                              your cameras abilities.



                              --- In atm_free@yahoogroups.com, "atmpob" <atmpob@y...> wrote:
                              >
                              > Can you explain a little more what you mean by that?
                              >
                              > Dale Eason
                              >
                              > --- In atm_free@yahoogroups.com, "jaf4oct" <jaf4oct@y...> wrote:
                              > >
                              > > Since
                              > > most Robo testing does not use the full intensity range of the
                              > camera
                            • jaf4oct
                              Scott, I really do wish I had a direct path back to that document. Unfortunately my encounter with it occurred long before the Internet. A man at Glass Fab
                              Message 14 of 20 , Feb 5, 2005
                                Scott,

                                I really do wish I had a direct path back to that document.
                                Unfortunately my encounter with it occurred long before the Internet.
                                A man at Glass Fab provided it to the company I worked for. Glass Fab
                                is also located in Rochester NY and at the time had strong ties with
                                Kodak. He set numerous items that he thought we would find interesting.

                                First I do wish to clarify I am not trying to start another hear say
                                Internet urban legend here. In fact my intent has little to do with
                                Foucault testing at all. I just wanted to keep people aware that the
                                level of testing accuracy obtainable on small optics does not
                                necessarily scale to larger ones.

                                With that said lets look a little closer at what I remember of the
                                comparison study. The student who performed the work did not aim his
                                research toward telescope making but instead was performing a study on
                                how the Foucault test could be used to test other conic constants
                                other than the historical ones in common use. This was driven mostly
                                by the new wave of PC based optical design software just coming into
                                vogue. Aspherics were a new hot item in lens design at this time and
                                he was showing what could and could not be made using existing equipment.

                                His test setup was very creative and he reduced his variables to the
                                minimum. He differed from your testing methods in that he used only
                                one mirror that was about a 12 inch near f3. He or a cohort at Kodak
                                figured this mirror to various conic constants between +4 and -4.
                                Since the radius was maintained other factors like source to mirror,
                                source to knife-edge proportionality, etc were maintained. Since the
                                Foucault test only deals with longitudinal aberrations by changing the
                                conic constant he was able to simulate many different optical
                                configuration with one standard setup.

                                To me his findings made perfect sense. As the total deformation on
                                the mirror increased, the width of the annulus at focus at any one
                                given knife position decreases. It is just like how the depth of focus
                                decreases in a microscope as you raise the magnification. When the
                                annulus becomes too narrow it will be difficult to judge when any
                                particular zone is at focus and errors will increase.

                                Doubling the aperture of a newt, keeping the F/ratio the same, also
                                doubles the amount of deformation required to parabolize it. In your
                                testing it looks like you are covering a range of about 8 waves of
                                deformation. The RIT study covered + and - 80 waves. This goes back to
                                my original intent, that you should not expect the testing sensitivity
                                you achieve on small optics to directly relate to larger ones.

                                JAF


                                --- In atm_free@yahoogroups.com, "starzkey" <starzkey@c...> wrote:


                                >
                                > >
                                > > RIT did a study in the '80s comparing various interferometers and
                                > the
                                > > Foucault test. The end results found that a Foucault test was
                                > accurate
                                > > to about 1-2% of the total correction.
                                >
                                > By any chance, is there a published, written summary of this study
                                > that you could point me to? Having recently completed a Foucault-
                                > interferometry comparo on several different mirrors ranging in
                                > asphericity from an 11" F3.6 to a 10" F5, I would be interested to
                                > read other's analysis and 7 description of the similarities and
                                > differences seen between these two methods.
                                >
                                > Scott Milligan
                              • atmpob
                                Thanks for the explanation. Actually it is not a harsh contrast. In fact I see more subtle variations using the camera than with my eye. To get harsh
                                Message 15 of 20 , Feb 5, 2005
                                  Thanks for the explanation.
                                  Actually it is not a harsh contrast. In fact I see more subtle
                                  variations using the camera than with my eye. To get harsh contrast
                                  I have to turn the contrast up past its normal setting. But even
                                  then it actually makes the robo test more sensitive because robo
                                  testing is looking for intensity matches. When the contrast is
                                  high it exaggerates the difference and narrows the zone of nulling.
                                  Making it more exact. We don't need the full range of the camera so
                                  as your original post stated robo does not make use of the full
                                  range. But I wasn't sure if you knew that it did not need the full
                                  range and was a limitation the did not matter.


                                  My robo also looks at intensity differences and for the 2 web cams I
                                  have tried I can show that there is a linier relationship that holds
                                  within the middle 1/2 of the intensity range. It seems to produce
                                  reliable results. At the same time I can compare the images taken
                                  by it with what I see behind the knife. There are very close to the
                                  same given all the other variables like display gamma and human
                                  perception mechanisms.

                                  Now if we were doing photometry that would be different.
                                  Dale Eason

                                  --- In atm_free@yahoogroups.com, "jaf4oct" <jaf4oct@y...> wrote:
                                  >
                                  > Dale,
                                  >
                                  >
                                  > Webcams were designed to produce acceptable video over a large
                                  range
                                  > of applications. For most users, scenes with harsh contrast (like
                                  > mirror testing) were to be avoided. To achieve this many webcams
                                  will
                                  > have a white clip and black clip. Clipping will reduce the full
                                  range
                                  > of the camera. The cameras DSP usually controls the clipping level.
                                  > You may think you have some control over this by the camera
                                  controls
                                  > provided by the camera manufacturer. These controls only affect the
                                  > DSP output not how the DSP will interpret incoming data. Camera
                                  > manufacturers have done a great job in finding ways to have cameras
                                  > mimic the human eye since that is what most users expect from their
                                  > cameras. Unfortunately the human eye is a poor photometer. Besides
                                  > clipping the video AGC and the camera gamma may not actually be
                                  under
                                  > your direct control as you may be only setting a range limitation.
                                  > Since CCDs are analog devices you are also affected by the A/D
                                  > converters, which may have their own logic adding unknown varibles.
                                  > Any of these effects can add non-linearity in the way the camera
                                  will
                                  > perceive the knife-edge.
                                  >
                                  > There are so many webcams that reverse engineering them is not a
                                  > practical endeavor. The only real choice remaining if you want to
                                  > understand your particular camera limitations is to characterize
                                  it. I
                                  > would suggest getting a Kodak Q-60 test chart. The Q-60 has a nice
                                  > linear step grayscale from white to black. Don't touch the chart
                                  with
                                  > your fingers since body oils can change the gray step relationship.
                                  > You may be surprised on just how poorly your cameras live video
                                  > functions on this gray step chart over varying light intensity
                                  levels.
                                  > It will however allow you to optimize your setup to make the most
                                  of
                                  > your cameras abilities.
                                  >
                                  >
                                  >
                                  > --- In atm_free@yahoogroups.com, "atmpob" <atmpob@y...> wrote:
                                  > >
                                  > > Can you explain a little more what you mean by that?
                                  > >
                                  > > Dale Eason
                                  > >
                                  > > --- In atm_free@yahoogroups.com, "jaf4oct" <jaf4oct@y...> wrote:
                                  > > >
                                  > > > Since
                                  > > > most Robo testing does not use the full intensity range of the
                                  > > camera
                                • Bob May
                                  If you can get a copy of that report, I can retype it into HTML and cut out the photos from the images of the pages. It would be a nice thing to have on the
                                  Message 16 of 20 , Feb 5, 2005
                                    If you can get a copy of that report, I can retype it into HTML and cut out
                                    the photos from the images of the pages. It would be a nice thing to have
                                    on the web!
                                    Bob May
                                    bobmay@...
                                    http://nav.to/bobmay
                                    http://bobmay.astronomy.net
                                  • Bob May
                                    Yep, there are indeed only 5 levels of gray that are transmitted. However, the decision to go to 5 levels is done at the output of the camera, not at the A/D
                                    Message 17 of 20 , Feb 5, 2005
                                      Yep, there are indeed only 5 levels of gray that are transmitted. However,
                                      the decision to go to 5 levels is done at the output of the camera, not at
                                      the A/D convertor. This allows the camera to give a decent range of shades
                                      when the contrast is jacked up.
                                      I jack up the signal level on my monitor to about 3 times the correct value
                                      and feed that into the monitor and that plus the monitor being at max
                                      contrast, I get a real diffence between white and black that really allows
                                      for the comparison of the sides of the zone at a mid level of gray on the
                                      monitor.
                                      I'll note that when you do this to a webcam, you don't see the step changes
                                      in the brightness and this tends to indicate that the levels are not being
                                      cut too early in the DSP of the camera.
                                      When I was working on the JPEG compression standard, 5 levels would
                                      sometimes produce a banding effect on gradually shading areas of a photo.
                                      When you went to 8 levels, allmost all of that shading went away (the
                                      detection level of the shading was highly person dependent!) so I do tend to
                                      agree with you on the subject of 5 levels being only minimally acceptable
                                      for visual use. The thing here is that the two ajoining levels had to be
                                      next to each other making a definite line that wandered about to see the
                                      changes. The thing tho is that on the Foucault test, the eye is hard put to
                                      get to that level of accuracy in measuring the gray of the zone openings.
                                      You really need to see the two different shades next to each other to see
                                      that they really are different. Robofoucault does a better job at this as
                                      it is measuring the count rather than trying to compare two shades of gray
                                      that are apart. This is why I like to have a light in the room illuminating
                                      the mask so that you compare each of the zones against the mask brightness
                                      as this makes it a lot easier to see that little bit of difference as you
                                      compare the zone to the mask on each side and, if the mask is evenly
                                      illuminated, the two sides should be in good agreement. I'll also note that
                                      if you have a systematic error in judging the shades, that will come out
                                      when you do the math as all of the zones will be off by the same amount.
                                      Bob May
                                      bobmay@...
                                      http://nav.to/bobmay
                                      http://bobmay.astronomy.net
                                    • Ken Hunter
                                      However... The left side and right side of the focault image are affected by the DSP equally. A brightness level on one side of the focault test will match the
                                      Message 18 of 20 , Feb 6, 2005
                                        However... The left side and right side of the focault image are
                                        affected by the DSP equally. A brightness level on one side of the
                                        focault test will match the same brightness level on the other and
                                        that is what the focault is doing. Matching brightness levels. The
                                        exact value may be misrepresented by the internal processing but the
                                        levels can be matched across the CCD.

                                        Ken Hunter
                                      • Bob May
                                        Yes, the DSP does tend to do this but the problem area here is that there aren t enough brightness levels to make the observation accurate. Bob May
                                        Message 19 of 20 , Feb 6, 2005
                                          Yes, the DSP does tend to do this but the problem area here is that there
                                          aren't enough brightness levels to make the observation accurate.
                                          Bob May
                                          bobmay@...
                                          http://nav.to/bobmay
                                          http://bobmay.astronomy.net
                                        Your message has been successfully submitted and would be delivered to recipients shortly.