Murchie's Analysis of the Scratch

Based on Murchie's E-message of 11 April 1995

From: Scott Murchie

Date: 4/11/95

To: MSI-NIS team

Subject: more on the scratch

In an attempt to better characterize the effects of the scratch on the cover glass of the CCD, I've assembled some data examination.

The illustrative figures are in the directory /project/near/SDC/MSI/scratch on retro, as are raw files used for this purpose.

[NB: This html document only contains the processed data; if you want the raw data, get it from the preceding directory on retro. Also, the images are GIFs rather than TIFFs. I've kept the .tiff from Scott's original message so you can find these on retro if you need to.]

[NB: click on the icon-sized images to get the full-size images.]

Fig1

Fig1.tiff is a dark-corrected image of the scratch, taken at 100 ms exposure with the 500 nm test filter at -30C. It is an average of 16 frames; an average of 16 dark frames taken at the same temperature and exposure time has been removed. (20580169.FIT was subtracted from 20581459.FIT.) It is an 8-bit data scaled so that 0,255 -> 0,255.

Fig2

Fig2.tiff is a 700 nm/500 nm ratio image, with dark removed. It was calculated by:

(20582029.FIT-20580169.FIT)/(20581459.FIT-20580169.FIT)

and the DN's are coded such that 18.25,22.5 -> 0,250. As expected due to differential refraction, the scratch is accentuated but is less pronounced than in the raw image.

Fig3

Fig3.tiff is a 1000 nm/500 nm ratio image, with dark removed. It was calculated by:

(20582679.FIT-20580169.FIT)/(20581459.FIT-20580169.FIT)

and the DN's are coded such that 1.5,2.0 -> 0,250. The greater different in wavelengths than in the previous ratio makes the scratch more pronounced, this time more comparable in magnitude to the feature in the raw imagery. However I'd be more concerned about the point blemishes, which are larger in magnitude.

Fig4

Ed Hawkins has maintained that the illumination conditions in these tests are not what will be experienced for distant objects with the CCD in the imager. To assess whether the scratch would show up under more realistic conditions, fig4.tiff shows an image of a piece of graph paper with a hole in the middle, taken during the focus tests before the CCD was moved exactly to the focal plane (in other words it is a little out of focus, with a spot size of 2-3 pixels). The 12->8 bit conversion was done to stretch contrast, with 2560,4095 -> 0,250. Neither the scratch nor the point blemishes are readily visible. (The raw file is raf8s025.002.)

Fig5

Of course they would be hard to see anyway, so fig5.tiff shows a filtered image to which an edge enhancement (high-pass filtering) was done. The box size was 33 rows x 21 columns, to give an approximately square box in real space on the CCD. Here the 12->8 bit conversion was done simply to fit the data into the available space, with 0,500->0,250. The grid on the paper and the edge of the hole in the paper come out like gangbusters, but I can't see any evidence of any of the blemishes - but as Mark and Jim have rightly emphasized, they'd be expected to show up more in color ratio images and we just don't have the right data!

Fig6

For comparison of what the edge enhancement does, fig6.tiff has the same procedure to the file represented in fig1.tiff. The DN range was expanded in the 12->8 bit conversion, with 0,125 -> 0,250. This shows how strongly the edge-enhancement filter picks up rapid changes in DN, by accentuating the scratch and all the little point blemishes (some of which may not have shown up as readily before). The LACK of these kinds of things in fig5.tiff is consistent with what Ed Hawkins explains.

I also have some profiles across the dark-removed raw and ratio images to show the magnitude of the effect of the scratch. Each figure shows profiles along parts of 6 columns.

In the 500 nm image (binhexed Kaleidagraph figure brt_profile.hqx), the DN drops by 10-15% along the worse parts of the scratch.

In the 700 nm/500 nm ratio image (binhexed Kaleidagraph figure 7to5_profile.hqx), the scratch shows up as having a higher ratio because more of the shorter-wavelength light has been refracted out by the scratch. The magnitude of this effect on color ratio is of order 5%.

In the 1000 nm/500 nm ratio imaged (binhexed Kaleidagraph figure 10to5_profile.hqx), the scratch shows up as more pronounced because of the greater difference in wavelengths. The magnitude of this effect on color ratio is now of order 10%.

Ed Hawkins figures that the light passing through any point on the cover glass will be dispersed over something like 30 pixels (which dimension I'm not sure). If this is true, then the effects on color ratio will be of order 0.2-0.3%.

I can't put my 2 cents in as much as some others can, because I'm between a big rock and a very hard place. But it seems we'd have more actual data on the magnitude of this problem if we put the imager back into the OCF IMMEDIATELY for a new test containing color imagery of well-illuminated targets with appropriate color and brightness variations, to see what effect these blemishes have. Replacing the CCD would be a big hit in terms of calibration time, so the team should have adequate data before pursuing that course.