Processing Notes 2017

Jubilees Palimpsest Project

Contents
December 12, 2017
December 5, 2017
November 14, 2017
October 31, 2017
October 17, 2017
October 3, 2017
September 26, 2017
September 19, 2017
September 5, 2017
August 29, 2017
August 3, 2017
July 6, 2017
June 29, 2017
June 8, 2017
June 1, 2017
May 11, 2017
May 4, 2017
April 27, 2017
April 20, 2017
April 6, 2017
March 30, 2017
March 23, 2017
March 1, 2017

December 12, 2017

First attempt at Laplacian Eigenmaps on C73inf

• Page: Ambrosiana C73inf 124, http://palimpsest.stmarytx.edu/AmbrosianaArchive/Ambrosiana_C73inf/Ambrosiana_C73inf_124/
• Region: x,y,w,h=3173,4510,850,118 (x is pixels right from upper left, y is pixels down from upper left)
  • It has never been successfully read, although we have a guess at a letter and have enough context to verify or falsify a proposed reading.
  • PCA and Keith's "Royal Blue Fluorescence Divided by Transmissive" (Ruby) have not helped.
  • It is iron gall erased text, iron gall overtext, and iron gall tincture.
  • There is no damage to the parchment (such as holes and tears).
  • The erased text is minimally or not at all blocked by overtext.
  • The script is distinctive and clear when it is visible.
  • Traces of erased ink visible at x,y,w,h=3880,4587,27,3 and 3887,4533,13,10
  • Within that same stroke is a region the color of parchment but surrounded by traces of ink in a recognized stroke (i.e., we believe ink to have once been there even though no evidence is visible in natural color). That region is x,y,w,h=3883,4556,6,6.
• Result: nothing too promising in first volley

Next steps

• Try combining fluorescence, reflected, and transmitted light captures into one data set. Might need some calibration from spectralon but basically the same data type (16 bit).
• Schroedinger eigenmaps
• Slightly easier ROI to the right of the initial ROI: x,y,w,h=4023,4510,850,118, reads "UNTBE" and keeps going to the right "STIAE"
• Rather easy ROI on a different page:
  • Data: http://palimpsest.stmarytx.edu/AmbrosianaArchive/Ambrosiana_C73inf/Ambrosiana_C73inf_062/
  • ROI: x,y,w,h=659,1681,934,204 (from upper left)
  • Reads: ET INHAB
  • Preview: http://jubilees.stmarytx.edu/iiif/Ambrosiana_C73inf_062_Ac_00.jp2/659,1681,934,204/full/0/default.jpg
• We also have lots of other manuscripts we could try.
• At some point it may be helpful to get more people (Todd) trying things even if they don't have quite the expertise.
• At some point could confer with others experimenting with supervised processing (e.g., Bill Sellers, University of Manchester, Canonical Variates Analysis).

More background on non-linear transformations

• Linear (PCA, ICA) is like looking from a different direction. Non-linear looks for clusters similar in some way and makes them look more similar (and dissimilar things more distinct).
• Classification means the user asserts that there are, for example, five materials in the region of interest. The software then sorts all the data into five bins, most similar things going together.
• Despite frustration at the first volley the concept sounds extremely promising for manuscript study.

Progress toward expanding access

• A former student of Roger works at Huawei, which makes a camera phone with second panchromatic sensor. A phone and access to guts is forthcoming.
• IMLS proposal in the works to expand access to ENVI-like processing functionality for free to museums and libraries.
• Some discussion of VIPS and the NIP2 graphical user interface. Possible, not clear, that could surpass ImageJ as a backbone for cultural heritage image processing.

December 5, 2017

Last call of 2017

• December 12, usual logistics (1:30pm EST, Todd initiates over Skype)
• Dave Messinger will join for the first half hour to talk about non-linear transformations (more below)
• Sarah Baribeau invited if she wants to hear about non-linear transformations, but not sure we will have broad topics of discussion for Spectral RTI by then (see below).

Plan for spring 2018

• Frequency: continue with weekly and cancel when necessary.
• Day and time: Thursday, 2:30-3:30 Eastern.
• Starting January 4, skip January 11, resume January 18.
• Scope: Mixture of general, Milan, and Berlin, but separate call for CCR.
• Lead: Josephine will take lead on agenda, Skype initiation, and notes.
• People: the usual and add Ted, Loren, and Ira (realistically only expect Loren and Ira when something important concerns them... when they are available adjust priorities accordingly).

Update on registration correction for filter magnification

• Bugs being squashed, such as certain fluorescence properties producing a functionally negative image which requires compensation in order for registration to work.
• Timeline hard to estimate but a breakthrough could happen quickly.

Questions about non-linear transformations

• Di Bai and Leidy did some work on Origen's Hexapla.
• The results were an interesting start but not quite successful yet.
• Only fluorescence images were used: not sure if that was an informed judgement or confusion about the data available.
  • Dave Messinger clarified that processing time increases with size of ROI, not number of images available.
  • The Ruby processing found good information in fluorescence so it may have been deliberate.
• Todd will follow up with Dave Messinger, Leidy, and Di Bai about the work they did so far and possibilities for what comes next.
  • Shroedinger has an advantage over Laplacian if certain pixels can be identified by the human as a target material (i.e., erased ink), either because it is partly visible or expected from rest of the stroke (make the distinction clear). The more the better.
  • Todd will pick some small and difficult regions of interest less than a 100,000 pixels for which linear transformations have not been helpful.
  • Square rather than cursive, iron-gall reagent rather than Gioberti.
  • Looking forward to the day when reasonably technical scholars can perform non-linear transformations themselves (Python, MatLab, Octave).

Athens

• Mike and others reported on general success in Athens
• Web publication cleared to proceed.
• Diplomatic progress may open possibilities for sustainable work along the lines discussed for northern Italy.

SpectralRTI

• Todd and Sarah will follow up about matters of troubleshooting.
• For the larger group there may be broader questions to discuss in the future.
• When is the cost in capture time and equipment justified?
• Shall we continue to use flattened hemisphere captures? (When looking at the broad image flattening has the uncanny effect of not making the "far" side darker. However, when one is zoomed in to one side, non-flattened source images just make the whole image darker or lighter as the virtual light position changes).
• Shall we continue to use images impacted by copy stand or other shadows? (More data has certain clear advantages but can add eerie effects.)
• Shall we continue to use gamma-corrected hemisphere captures? (It makes sense since all other RTI does so, but I'm not sure a comparison has been completed.)
• Shall we continue to use Photoshoot (which uses a LUT) rather than ImageJ (which uses an equation) for gamma correction? (I'm already assuming that gamma-corrected reflectance and fluorescence images needs not be treated as archival.)

November 14, 2017

Today's processing call was mostly concerned with preparing for the Society of Biblical Literature in Boston (where Mike, Keith, Roger, and I will be presenting), and the subsequent conference in Athens (including Ken, Keith, Roger, and Mike).

Keith's registration correction for magnification due to filters is still in debugging mode and will not resume until after Athens. We discussed the baseline to which other images are corrected. For Keith's processing it is the main bank 365 nanometer reflectance because it tends to have the most detail. The hemisphere captures should be fine without any registration correction because they are visible spectrum only, heavy on the blue (since they are white LED). Some of the images Roger created with the opposite side of the folio registered to the first side are cropped. It sounds like a simple transposition (without rotation) should suffice to align that data with the hemisphere captures, accurate color, etc. Roger may have a way to avoid cropping in the first place.

Once we have registration corrected data for filter magnification and for opposite side it may be worth another push to non-linear transformations.

The next call will be December 5 at the normal time and medium. Mike and Damian expect to be able to join from Sinai, perhaps using cell phone network. We'll reevaluate the time for 2018.

October 31, 2017

Today's processing call included Keith, Roger, and Todd. We're going to sit out next week and resume at the usual time on November 13 (after daylight savings). Here are some notes:

Opposite side of folio

• Roger demonstrated ability to render both sides in different colors in the same image. (Amb_C73inf_123_flipH_warp_124_cal_b01-b14_28bands_RF_R7-123_G7-124+B1-124_subs50.jpg)
• Todd thinks that simple color or a single reflective visible of the opposite side flipped and warped would do the trick. If they are aligned it would be easy to flicker between them to identify the layer of writing in which a stroke originates.

SBL

• Mike will talk about Sinai, Todd will talk about Milan, the only real coordination necessary is Roger and Keith. They will talk Nov. 6, 1:00 EST.
• We are trying to hold Monday evening for dinner with potential collaborators (even though none have been identified yet).

2018 Scholars Workshop

• Todd still hasn’t been able to firm up March 12-15. Backup options might include attaching the scholars workshop to something else, such as R-CHIVE.

Registration correction

• Applies especially to UV Block and Pass, but to some degree to all captures from the main bank reflectance and auxiliary fluorescence (not hemisphere captures... guessing not transmissive)
• (34 fluorescence images/page + 14 reflectance images/page) * 144 pages in C73inf * 100 MB/page = 675 GB just for C73inf
• (24 fluorescence images/page + 14 reflectance images/page) * 44 pages after C73inf * 100 MB/page = 164 GB after C73inf
• Rather than processing and uploading all images, Keith will share the software with Todd who will run it on an attached computing cluster. A few pages will be verified against images created by Keith.
• We will start with C73inf and think about the supplemental palimpsests in the future.
• If the data is ready by SBL Todd will share it on physical media in Boston, otherwise postal service.

October 17, 2017

Show-through from opposite side issue in general

• Rough transformation of opposite page as a user interface issue for human consumption... Todd will explore in IIIF Presentation and Image APIs
• Ruby shows text from other side as a distinct color... sometimes, not always clear
• ENVI can do high quality transformation ready for statistical analysis
  • Involves 162 points manually identified, especially edges and pin pricks (not text, which blurs coming through parchment)
• Hoku has some such ability from Livingstone data, not a higher priority than filter distortion
  • A different kind of warp compared to filter distortion, which has a center and uniform degree of magnification

What's next with c73inf

• Some alignment work done on pages 112-111 in Milan, more work could be done
• Todd will identify another folio of high scholarly significance

Next R-CHIVE working date is June 7-8, 2018

Milan summer 2018?

• Won't hear until end of March, not much time to plan before Ambrosiana's summer break.
• The point is to establish a long-term sustainable project, could benefit from thorough planning.
• September or January (again) might be an option, maybe even summer 2019 if NEH agrees.

SBL, Boston, November 20, 9:00 am

• Idea: Roger could do more history of discipline (like at R-CHIVE), Keith more current image processing
• Registration correction... more progress after next Tuesday

Next call will be October 31 (skipping 10/24)

October 3, 2017

Hoku

• Pseudocolor in Hoku
  • The Pseudocolor available now in Hoku is an early version, not the latest version
  • The difference is that the early version has a veil problem (appears around high contrast areas)
  • On the roadmap to include in Hoku the method use an initial mask of high-contrast area
  • When Hoku asks for a width and height, it means the tile size for processing samples, defaults to width,height=501,501
  • The Pseudocolor formula is one one that could be easily recreated for ImageJ or other.
• Improvements based on issues that came up at Manusciences
  • Preserve metadata except for dng tags, which confuse MacOS Preview
  • Increasing tolerance for spaces in file paths
  • Spectral angle mapping working
• The ROI option for Hoku Pseudocolor PCA is a placeholder not yet implemented
• Hoku is planned to be available on GitHub

Registration correction

• In the debugging phase, usable data within a couple of weeks
• Worth waiting for corrected data for C73inf

S36sup

• Todd’s observations in the processing guide are not surprising (clipping where stroke fades into parchment, convergence of output with KTK Pseudocolor)
• Show-through is concern here, more so in C73inf (see next)
• Plugging away at S36sup is not the best use of time

Show-through

• Roger has a good student working (Jackson Nappon sp?) on tools to align opposite sides of a folio using transmissive data
• He might benefit from participating in these calls, if his schedule permits
• Todd asked about approaches focusing on creating new 2D images vs. visualization tools using 3D engines

Miscellaneous

• RIT just inaugurated a new president, festivities included various interesting conversations about exciting possibilities in Croatia (near northern Italy), India, Dubai.
• Conferences coming up: SBL in Boston, Athens the following week

We will speak again at the usual time one way or the other. The agenda may shift toward topics from Mike.

September 26, 2017

We had a light day on the processing call, partly because Keith and I have been otherwise occupied lately. The progress was from Roger's work on Ambrosiana_S36sup. It is an odd specimen. Reading the undertext seems to go quickly from very easy to very hard. Rubrication may explain some of this, but not all. Separation is a challenge as well. Spectral angle maps can be helpful as an enhancement when there is a lead, but less so for pure discovery where nothing can be seen. It is worthwhile to run through advanced processing for the remaining pages, but not to get too bogged down if success is not forthcoming. Once we round out our tour of the supplemental samples we can return to the flagship, C73inf, equipped with lessons learned and registration-corrected data.

September 19, 2017

We had an extensive processing call today driven by Keith’s development of Hoku and news from Manusciences in Nice, France. My notes are below. Next week we will continue the regular schedule and will almost certainly have more questions for Keith about Hoku.

Ambrosiana S36sup (Gothic Bible)

• The processing guide gives some codicological information and the expected transcription for each page. http://palimpsest.stmarytx.edu/AmbrosianaArchive/Guides/Ambrosiana_S36sup_ProcessingNotes.html
• The first page has a region interest defined where past editors did not propose text. However, because it is a biblical manuscript editors never lacked a basis for guessing. Their guesses should be re-examined. Other scholars may also suggest ROIs, and Todd will add more as well.
• The last page is believed blank. Probably right, but would be a cool long-shot to find something.
• Todd will add a paleography chart to the processing guide.
• We are NOT waiting for registration corrected data for S36sup, but do hope to have registration corrected data before returning to our flagship, C73inf. Keith estimates that will take a couple of weeks.

Update on affordable ENVI

• Jessica Rogers had a brief conversation, will travel to Colorado for a fuller conversation

Report and miscellanea from Manusciences

• 40 students, 10-12 from US, 1 from Ethiopia, rest from Europe
• Interest from Manusciences in event at RIT in spring or summer, alternating years, will be discussed at R-CHIVE meeting this Friday.
• Josephine attended multiple practical workshops, report sense of variety of tools available to pursue different questions that can be asked.
• Daniel Stökl is doing philological work with Matlab

Demo from Keith on Hoku

• Alpha development jar and documentation available at http://www.cis.rit.edu/~ktkpci/hoku.html
• Demo of core functionality (specify directories, shelf, desktop, readtiff, pack, show, colors and line distinguishing required and optional specification, rubber-band boxing to switch from edit to execution mode, see documentation above)
• On horizon: write stack, registration correction, looping procedures, fast ica (uses integers rather than floating point, to be confirmed), improved documentation, better support for reading and writing ENVI files, publication on GitHub once further along in development
• Will Hoku ever replace ENVI lite? Not really because batch focused, not interactive
• Todd is interested in making ImageJ2 and Hoku play well together.
• Not designed to run headless (without GUI) but could be made so.
• Time exhausted before questions… will have more next week

September 5, 2017

Registration correction and supervised processing plans

• A sharable draft of the software should be ready at Manusciences.
• Registration corrected data for S36sup and C73inf should be available at the beginning of October.
• Todd will have scholarly notes on regions of interest by then.
• Then supervised processing will focus on finishing up our extended samples with S36sup, then return to the flagship of C73inf.

Schedule adjustments

• Manusciences will conflict with the regular schedule of 1:30 EST next Tuesday, Sept. 12.
• Thursday, September 14 at 2:30 pm EST (8:30 France, 1:30 Central) should work.
• Keep agenda flexible for questions that may come up from students or updates to Todd and Mike back in U.S.
• We will be back to our normal schedule in two weeks (September 19).

Processing software options and long-term planning

• ENVI will be available to students at Manusciences, but not afterwards (unless their institutions have it anyway). There is a concern about teaching people to process with tools they will never use again.
• Jessica Rogers is talking to the makers of ENVI about a stripped-down version for a more reasonable price. Guesses at “reasonable” start at $1000.
• Pre-proposal to use IMLS money to write free software to do cultural heritage processing. Several years out at best.
• Keith’s software will be called Hoku (the Hawaiian word for “star”). Has its own GUI, manages memory well, can read some ENVI files, has PCA implemented for whole image (not ROI), will eventually have capability for supervised processing.
• The SpectralRTI_Toolkit is being rewritten from an ImageJ macro to an ImageJ2 Java plugin. ImageJ loads entire images into memory before processing.
• We should keep an eye out for opportunities to focus efforts, avoid duplication, and minimize incompatibilities.

August 29, 2017

Update from Keith on registration correction software

• With some help from Ivan C., will be available for Mac, Windows, and Linux
• Time goal is to have it ready for alpha distribution at Manusciences, with the idea that a patient user can apply independently (as opposed to sending data to Keith for processing)
• There was a recent development that improved processing time from over an hour per page. A new method works better and faster.
• Todd offered to use a high-powered computing cluster available on his campus should an occasion arise in which heavier processing would be beneficial.

Update on C73inf

• High-school interns worked on the first seven pages of C73inf. The processing was sometimes helpful but did not suffice to make the unreadable portions readable, mostly the portions originally written in red ink. Specific regions and detail images are available in the processing notes, http://palimpsest.stmarytx.edu/AmbrosianaArchive/Guides/Ambrosiana_C73inf_ProcessingNotes.html
• Some of the images were uploaded as jpegs. Roger is looking for the tiff files.

Update on S36sup (Gothic Bible)

• Basic images are in the IIIF repository (http://jubilees.stmarytx.edu/iiifp/)
• The properties of this manuscript are unusual. The erased text is sometimes darker than the overtext, yet in some places there is no text visible where one would expect. The original ink corroded the parchment.
• Todd is out of his element on Gothic paleography, will solicit assistance.
• Roger is unlikely to have time to work on this before Manusciences. Right now Roger’s only help is from an undergraduate (David Lewis). Tania is also around and may have some interest in helping with some image processing despite major duties elsewhere (grant writing).

R-CHIVE call on Friday, September 8, 9am EDT

• Josephine inquired as to whether Keith and Todd are invited
• Will focus on possible funding partners
• Preliminary proposal to IMLS this week about getting tools into the hands of librarians, which could mean viewing options or could mean processing tools such as key utilities in ENVI rewritten in Python (an alternative is to get the makers of ENVI to offer a stripped-down version)

Update on pan-sharpening as a way to deliberately reduce exposure time or increase depth of field, or correct depth-of-field focus issues after the fact

• Roger and Nicole made an initial attempt which failed
• Dave Messinger thought the problem was a data mismatch. Pan sharpening uses information from all wavelengths to sharpen individual images, but jumping from reflectance properties to fluorescence properties is different.
• Roger hasn’t given up, but tabled for now

Resources for people at Manusciences to work on the Milan data (not all this was discussed in the call but was requested by Keith and I thought I would share with the whole team)

• The IIIF repository is online at http://jubilees.stmarytx.edu/iiifp/
• The data archive is online at http://palimpsest.stmarytx.edu/AmbrosianaArchive/
• The processing notes are online at http://palimpsest.stmarytx.edu/AmbrosianaArchive/Guides/
• Wide-audience updates are posted in the news section of http://jubilees.stmarytx.edu and http://twitter.com/thanneken
• C73inf is distinctive in that it we captured all preserved pages and we have captures of transmitted light
• C73inf includes Latin Moses (Jubilees and Testament of Moses), which is in two columns and harder to read (see the processing guide for the hardest-to-read regions)
• C73inf also includes the Latin/Arian Commentary on Luke, which is in one column and easier to read except for the rubrics (see processing guide)
• The major challenge with C73inf is the reagent. The Ruby images worked better than the Pseudocolor and Sharpie images.
• A79inf (Petrarch) is the only example of graphic art that we captured. The core question is whether the book in the image has writing and whether that cryptic writing can be deciphered.
• F130sup (Greek Commentary on Luke) includes a variety of tinctures and lack thereof.
• H190inf is an unidentified palimpsest. Roger’s images help but more work will probably be necessary to arrive at an identification.
• O39sup (Origen’s Hexapla) uses different kinds of tincture, includes occasions when it is not slathered on with a brush but “written” on with a pen, creating the illusion of an ink stroke where there probably had not been one. Nicole Polglaze had success with the tinctures on these pages.
• S36sup (Gothic Bible) has not undergone advanced processing yet. The script and the properties of the ink are unusual.
• All images having raking light and WebRTI versions. Because these palimpsests were heavily handled and often treated with reagent, we have not (yet) repeated the success of 2013 in reading the corrosion of ink into parchment where ink traces are not visible with spectral imaging. However, we can see scribal scoring lines that guide us to where text should be and generally give a feel for the condition of the parchment.

Update on calendar

• Next week we have a regular call, same basic format
• In two weeks several people will be in France for Manusciences. We will hold the appointment for now but remain flexible. It could be cancelled. It could be used to update folks not at Manusciences about Manusciences. It could be used for people at Manusciences to talk to Todd and Mike about the Milan data.
• Most of us will be at SBL Boston in November. We can share our papers and slides in advance to minimize redundancy. We discussed the fact that SBL differs from the sciences in that papers are usually read verbatim. It is still okay to narrate a slide as long another mechanism is in place to align the timing precisely.
• March 2018 at NYU is still in light pencil because our host just started a new job.
• Returning to Milan as early as summer 2018 is still on the table. We will hear from the NEH in March but are not putting all our hopes in the NEH.

August 3, 2017

Fall schedule

• Starting August 29, we will talk every Tuesday 1:30–2:30 eastern time (which will change for Keith when DST ends).
• Todd and Josephine will coordinate agenda items and efficiency so that most weeks there will not be need for a second call with mostly the same people. The calls will not all be exclusively dedicated to Milan or any other one project. The participant list may expand or contract on a week-to-week basis, but the core regulars should be able to hold Tuesday 1:30-2:30.
• We will not have another full call in August before August 29. We all have our homework (Roger is working on S36sup [Gothic Bible], Keith is working on the registration correction software).
• The long-term plan for Milan is to finish the last of the supplemental samples (S36sup) before mid-September (Manusciences). By that time registration corrected data (or the tool for correcting registration) will be available. With that and other lessons learned we will return to our flagship, C73inf (the Jubilees Palimpsest).

Using pan-sharpening to correct for out-of-focus or low-resolution captures

• The issue that precipitated the discussion is that Roger observed fuzziness in the fluorescence captures of H190inf at the top of the page (which is the outermost part).
• The metadata confirms that those fluorescence captures have aperture settings as wide as 4.3. The reflectance captures have aperture settings at 8.6. Thus it makes sense that the depth of field is narrower for the fluorescence captures and the edges would be out of focus.
• Depth of field is a general concern (and came up recently when Greg was interested in doing focus stacks).
• Dave Messinger has been working on pan-sharpening, which is using a high-resolution capture to sharpen a low-resolution capture. His use case (or one of his use cases) is to sharpen low-resolution hyper-spectral data.
• With the data we have it might be possible to use the tight-aperture reflectance captures to sharpen the focus on the wide-aperture fluorescence captures (at least when the visible features are basically the same).
• Another approach would take place at capture. It would be possible to deliberately bin four capture pixels into one so that the sensitivity of the bin would be (roughly) four times greater and the exposure settings could adjust accordingly (shorten exposure time or tighten aperture). The data coming out would have lower spatial resolution but better signal to noise ratio. The data could be pan-sharpened later.
• All of these are ideas with which to experiment. It remains to be established that there will be a substantial net benefit from all the manipulation and processing, as opposed to say just lengthening the exposure or putting up with some fuzziness.
• A core question stands: How much high-resolution data do we need to capture to sharpen data that has a lower spatial resolution or poorer focus for reasons of depth of field or shape of object?

Scholars meeting around March

• We are at a very early stage of planning a meeting of scholars to learn the benefits of the Jubilees Palimpsest Project (the data, processed enhancements, viewers, annotation and collaboration tools). The scholars will also offer feedback about what could be better.
• We don’t have a huge budget to bring in all the scientists we might like, but everyone is more than welcome.
• The most likely venue is New York University. Second most likely is Notre Dame.
• The first set of dates floated so far is March 12-15.
• Stay tuned for updates.

H190inf (unidentified erased texts)

• We didn’t discuss every interesting feature of H190inf. There is evidence of some double palimpsesting (or ink bleed) and a “touch up” of the overtext rubric in an ink that looks close to the eye but different in spectral signature. http://palimpsest.stmarytx. edu/AmbrosianaArchive/Guides/ Ambrosiana_H190inf_ ProcessingNotes.html
• I (Todd) will ask some scholars to see if they can identify the undertext(s) based on the evidence we recovered.

Nicole is ending her program at RIT with a big presentation tomorrow. We are all grateful for her work and hope she can remain involved in our projects and the field to the extent most beneficial for her.

Follow-up

From: Damianos Kasotakis

The aperture that we use doesn't drop bellow f5.6 (Ken could confirm or correct me on this) yet again this is pretty wide open aperture, when compered to f8. I guess it could be wrong value reported by Photoshoot? Or vise versa, a wrong value that appears on Photoshoot as f5.6, when in reality its closer to f4.3.

Focus stacking is a very nice technique, but is also very much time consuming. Considering a project where you already have 100+shots and 15-20 min of exposures in total.

A more realistic approach would be to shoot everything with a higher aperture (f11-f16?) and get STRONGER lights (yes Ken, MORE LIGHT) to cut down the exposures or make them equal as now, since we will be getting less light with that narrower f11 . There is no way that in that aperture you would have any kind of focusing problems. During our project in OKC, we were almost consistent with an f8.5 but we had longer exposures since the lights had to be moved way back for the RTI.

From: Ken Boydston

More Light! Indeed.

We doubled again the 365nm in both Auxiliaries and Mains, and increased by 50% 385 and 400 in the Auxiliaries. If the Auxiliaries are positioned optimally, the benefit is significant. We did not quite have time to implement the new Auxiliaries at MOTB. Optimal position and new lights would have made nearly 2 stop improvement at MOTB.

PhotoShoot includes a feature we should put to use. There is button on the main window which reports "100". This is the ISO setting Push this button once, it goes to 200. Push it again, it goes to 400. Push it again, back to 100. . It could be set always to 400; the increase in noise for the short shots would be just noticeable on close inspection. I would shoot the short shots at 100, long shots at 400. The ISO button is not active during pause, so long shot shot table should not be linked to short shot table. 10 second exposures at ISO400 would be a lot quicker and produce better results than 30 second exposures at ISO100.

With new more powerful lights and 4X ISO, we should have much improved images with much shorter exposures.

Maybe "pan-sharpening, which is using a high-resolution capture to sharpen a low-resolution capture" could be implemented to possible benefit in a couple scenarios:

1. Bin a high res chip for the long exposure shots. This results in lower resolution image with correlated increase in sensitivity and faster captures. The lower res image might be easily interpolated for use with a higher res luminescence-like image.
2. Capture a [functionally] low res image with lots of bands (maybe mostly out of focus), then focus stack using a high res capture of only luminescence (or rgb) at each of the focus planes, then combine.

BTW, I think max aperture is about 4.8; if shutter is programmed correctly, correct values are reported in PhotoShoot.

July 6, 2017

Embossing effect

• Roger noted embossing effect in this image: http://jubilees.stmarytx.edu/ iiif/Ambrosiana_O39sup_081_ NAP02_00.jp2/1681,3371,866, 676/full/0/default.jpg
• The filename suggests it uses reflective bands, not the fluorescence filters.
• Keith had indicated that even the reflective bands (with no filters at all) show some registration shift... is that what we are seeing?
• Todd noted that the embossing effect appears on the overtext and not the undertext, wondered if that is because the overtext appears in more reflective bands (especially IR) than the undertext.
• Noted to ask Keith about these questions.

Ambrosiana_O39sup

• With the possible exception of returning with spectral angle maps and Laplacian eigenmaps, we are done processing and posting O39sup.
• One intriguing discovery on page 81 indicates that letters that appear to be visible are actually strokes where Mercati "wrote" with the reagent where he expected to find a letter, as opposed to swabbing a general area. The evidence is described in the summary at the top of the processing notes (http://palimpsest.stmarytx. edu/AmbrosianaArchive/Guides/ Ambrosiana_O39sup_ ProcessingNotes.html and summarized here.
  • In Nicole's image (NAP02) the letters read by Mercati appear clearly: ιριβι http://jubilees.stmarytx.edu/ iiif/Ambrosiana_O39sup_081_ NAP02_00.jp2/1737,2247,366, 158/full/0/default.jpg
  • However, they are different colors, the rho and beta are dark red and the iotas are light blue.
  • The accurate color shows that those strokes are reagent (blue).
  • If they were reagent plus erased ink we would expect at least some reddishness in NAP02, as is the case in another example on the page: http://jubilees.stmarytx.edu/ iiif/Ambrosiana_O39sup_081_ NAP02_00.jp2/1885,3442,144, 112/full/0/default.jpg
  • The stroke we are seeing of the iotas, especially the second one, bear no trace of erased ink (dark red), only reagent (light blue).
  • Mercati wrote with reagent what he expected to see and reported in his edition that he saw it, but our imaging finds no trace of it (if there was ever ink there, it has been destroyed, probably by erasure)

Option for what's next: Ambrosiana_H190inf

• unknown Latin texts (plural because vary between perpendicular and parallel to upper text)
• good opportunity for discovery
• no reagent... a walk in the park or a challenge because erasure is so complete no one tried to read it yet?
• we have seven pages from seven folios (i.e., not both sides of any folio)

Option for what's next: Ambrosiana_S36sup

• undertext is a fourth century translation of Paul's letters (New Testament) into Gothic (an early relative of German)
• very significant for scholars even if few surprises because we have readings based on previous scholarship and comparison to other versions of Paul's letters
• iron gall reagent... we're starting to have some success with this and more success would be important progress
• we have twenty pages from ten folios (i.e., both sides of each folio)

Option for what's next: go back to Ambrosiana_C73inf

• This is still our banner exemplar and the only data set for which we have transmissive
• Further insights into defeating iron gall reagent should come back here eventually

Next week's call

• July 13 call moved to 8:00 pm CDT

June 29, 2017

Update from Keith on software correction of registration errors in filter captures

• As most of us saw in Rochester, the principles of calculation of center and degree of magnification are basically working and offer great hope for increasing usability of filtered images, especially those through glass filters.
• The latest area of troubleshooting is to eliminate outliers caused from failure to determine registration shift on materials other than parchment.
• Samples will be compared to the center of the image (by average spectrum vector). If they do not bear a sufficient resemblance to the center (presumably ink and parchment) they will be excluded from the data for calculating center and degree of magnification.
• Future development will allow the user to override the assumption that the center is typical of the object of interest (e.g., if the frame consists of fragments around but not on the center).
• The next few weeks will be occupied with other tasks (New Hampshire conference, Oklahoma City).
• Other questions arising from the conference will be addressed later.

Results of focused processing based on comments from Origen Hexapla scholar Ben Kantor (BPK).

• Nicole processed images for all the pages captured from Origen’s Hexapla (O39sup). Todd is catching up on processing and posting (currently the last two folios remain to be completed).
• The processing guide includes a round of feedback from Ben Kantor about especially crucial and unreadable regions. http://palimpsest.stmarytx. edu/AmbrosianaArchive/Guides/ Ambrosiana_O39sup_ ProcessingNotes.html
• The processed images, especially Nicole’s, are very successful at recovering text when the main obstacle is erasure.
• The processed images are also rather successful when the additional obstacle is overwriting.
• It is no surprise that we had no success when letters were physically cut away or covered with a heavy illumination.
• So far the processed images have not improved on accurate color when the obstacle is Gioberti Tincture (see next)

Gioberti Tincture

• Gioberti Tincture appears blue in the accurate color images
• We speculated that the intended chemical reaction loosens the traces of ink and, when applied too heavily, causes the soupy mixture to blob and bleed outward, leaving the chemicals visible but the shape of the original letter indistinct.
• See especially http://jubilees.stmarytx.edu/ iiif/Ambrosiana_O39sup_061_Ac_ 00.jp2/788,4712,1204,1368/ full/0/default.jpg
• For those of you not yet familiar with the IIIF Image API, that means image(owner_shelfmark_page_ AccurateColor_diffuselight)/x, y,w,h coordinates/full resolution/no rotation
• The linear transformations seem not to be helping for Gioberti Tincture, but the fact that we can make some out in accurate color means we’re not ready to give up.
• Nicole will try using spectral angle maps in ENVI
• Looks past darkness and brightness for similar spectral signature (difference between reddishness and blueishness, etc.)
• Similar materials have similar spectral vectors (a smaller angle between them)
• Todd will share the data and key regions with Di Bai for him to process, will also try to get Laplacian Eigenmaps running on his own systems.
• More optimistic about Laplacian eigenmaps working on Gioberti tincture than gallnut.
• Would be interesting to try XRF.

What’s next for processing after O39sup

• H190 inf, undertext is unknown Latin text, but we want it to be known
• S36 sup is a fourth-century translation of biblical books into Gothic… shows how Christianity was “translated” from a Greek-speaking Mediterranean context to Germanic tribes in northern Europe.

Plans for next two weeks

• Keith will be on an island out in the ocean (a different one) but Roger, Nicole, and Todd will have plenty to discuss. Others are welcome to join.
• In two weeks key personnel other than Todd will be in Oklahoma City. Tentative idea is to Skype at 8:00 pm CDT on Thursday July 13.

June 8, 2017

Keith’s work correcting registration errors from filters

• The attached file “Graph_Paper_vs_Image_Data. pdf” illustrates the challenges of measuring magnification from in images of real objects. The graph paper was fairly consistent but when applied to manuscripts (F130sup) the software has spikes of over- or under-estimating the registration shift.
• The next step is to identify and remove the spikes.
• 400 random spots measured across the frame with minimal overlap.
• Question: is there a relationship between the error spikes and physical material, such as parchment cockles or microfiber?
• Question: how will registration correction fit into the workflow and quality control? Possibly MegaVision Photoshoot software but more likely shortly downstream.
• The glass filters produce the most magnification (5-10 pixels at edge). The Wratten filters produce some (1-2 pixels). Also some in the mainbank across wavelengths, up to four pixels in IR.
  • Mainbank demagnification is a result of the lens.
  • Not necessarily centered if object plane and sensor plane not parallel.
  • In focus but has a magnification change.
  • Apochromatic lens we have is corrected for focus, perhaps not for magnification.

Nicole Polglaze has been doing some work on O39sup (Origen’s Hexapla) at RIT with Roger.

• Notes and some comparative detail images are in the processing guide: http://palimpsest.stmarytx. edu/AmbrosianaArchive/Guides/ Ambrosiana_O39sup_ ProcessingNotes.html
• The five columns are a transcription of the Hebrew Psalms (one Hebrew word per row) into Greek and four Greek translations.
• Reagent: successful processing of reagent on pages 1 and 27
  • Relationship to C73inf reagent? Is it a different chemical composition? Less thickly applied?
  • Are we learning lessons that can be taken back to C73inf?
  • Will our data with different manuscripts and reagents be of use to Greg’s work and eventually a typology of reagents and their properties?
  • Behavior of reagents over time?
• Blur and divide increases sharpness, does not reduce noise.
• Hue rotation: Nicole has been doing it, someday would like to provide option of hue rotation in viewing environments.
• KTK automated Pseudocolor working well in O39sup (didn’t in C73inf)
• Page 30 detail in the guide shows ghosting/shadowing effects when overtext written over undertext. Various related factors discussed (registration errors for filters, bleeding ink stain, physical shadow). Humans may be better than machines at ignoring ghosting.

Texture imaging (raking and RTI)

• Compared to the palimpsest in the 2013 phase, we are not seeing corrosion of ink into parchment as much in the Milan data, probably because Milan palimpsests were much more handled and “varnished” with reagent and attempts to conserve.
• Even if raking and RTI do not directly show text, they still provide a feel for the physicality of the object, its wrinkles, and condition with materials accreting on or eating into the surface.

R-CHIVE

• Most/all of us plus Josephine and hopefully Greg will be on a call Monday afternoon 2pm EDT to discuss point persons and pre-planning for sessions in Rochester.
• IR fluorescence could be another topic to explore in Rochester.
• Slots for 45-minutes presentations should include ten minutes or so for questions, roughly.

Next call

• Next Skype conference dedicated to Milan processing will take place in three weeks (June 29) at the usual time (2:00 pm EDT).
• Not much urgency on most questions, but if we have insights for Nicole bring them up while we are in Rochester.

June 1, 2017

Welcome Nicole

• REU student from Grinnell college, originating from Boulder
• nappci@rit.edu
• Processing options:
• C73inf for a challenge with reagent
• O39sup for new material with high potential for processing success and scholarly impact

Keith’s Registration Correction Software

• Keith will give a talk on registration correction at R-CHIVE
• Software in a functional state
• Quality better on UVP/UVB than Wratten
• Ability to detect a shift without noise is harder on manuscript than graph paper
• Error reduction refinements in progress, close to being done
• Block and pass easily uploaded, the rest will be lots of data... better idea in a week whether we should bring drives to Rochester or rely on Internet
• Processing time requirement: 20 minutes for main bank images, 2-4 minutes for filters (on old computer)
• SIFT for translation correction patented by a university, poor license
• Various options for Windows and Linux packages (Mike knows developers, Todd can test on Windows and Linux)

R-CHIVE

• Link to R-CHIVE conference registration: http://humanities.lib. rochester.edu/rchive/index. php/conference/
• Keith will invite Adrian Wisnicki and Fenella France

International programs

• Dave Messinger and Roger met with associate dean for international studies about Ira's idea of joint PhD between Rochester and Hamburg
• RIT campus in Croatia has affiliation with Erasmus program (EU collaboration with other places)
• NSF PIRE (partnerships for international research and education)
  • https://www.nsf.gov/funding/ pgm_summ.jsp?pims_id= 505038https://www.nsf.gov/ funding/pgm_summ.jsp?pims_id= 505038
  • 3-5 years several million, collaborations in US and overseas
  • ND or StMU could also be part of collaboration
  • ND has a “Global Gateway” in Rome that includes Milan
  • Perhaps also Russia

Next week continue on Skype at same time and format before in-person meeting in Rochester the following week.

May 11, 2017

Updated processing guides

• http://palimpsest.stmarytx. edu/AmbrosianaArchive/Guides/ Ambrosiana_C73inf_ ProcessingNotes.html#20170510
• http://palimpsest.stmarytx. edu/AmbrosianaArchive/Guides/ Ambrosiana_F130sup_ ProcessingNotes.html
• http://palimpsest.stmarytx. edu/AmbrosianaArchive/Guides/ Ambrosiana_O39sup_ ProcessingNotes.html This page is new and specifies which of the pages of Origen’s Hexapla are important and which are only imaged because the other side is important.
• The guides on http://palimpsest.stmarytx. edu/AmbrosianaArchive/Guides are also mirrored to http://jubilees.stmarytx.edu/ Guides/ since the former can be flakey with making connections at times.

Reagent

• Lots of work to get even moderate improvements to Accurate Color in C73inf (see bottom of reading guide above)
• The reagent in C73inf does not behave as expected from other experience
• Ira Rabin has worked on Tanic acid behavior (adheres to traces of ink and suppresses fluorescence, less so to parchment so ink is darker in typical fluorescence images), will be interested to see data from C73inf
• Greg Heyworth is working on reagents (and recreating them), will be interested to see data from C73inf
• The guides will be a resource because they indicate the filenames most successful, Roger will be equipped to notice a pattern (if there is one)

Better processing through alignment of both sides of transmissive

• The software Keith is developing to correct registration in filter images could be useful here as well
• Ink bleed through in Livingstone set precedent for identifying layers of ink on both sides
• C73inf has scribal punctures that should aid alignment (with some human supervision)
• Todd has some ideas for viewing environments that will approximate “real life” study of flipping sides of folio.
• A senior project at RIT on just this should be done exactly a year from now.

Better processing through PCA

• Roger has had success both with a small ROI of only erased ink and parchment, and with larger ROI including upper text. No simple preference.
• Roger now tends not to use more than eight bands (selected from appearance), but Todd’s big-stack approach could work too.
• Roger and Keith confirmed the validity of Todd’s work on running PCA on subsets of data, combining and culling the principal components, then running PCA again on the principal components.
• Comparable to, but not identical to minimum noise fraction, which culls low value principal components and rotates back (reverse eigen vector) to image space for a reduced-noise image on which PCA can be run again.
• Comparing inverse PCA to progressive PCA (PCA on PCs) could be another senior project at RIT
• Sequence of stack doesn’t matter (determines eigen vectors but not components)

Better processing through ICA

• Roger still uses PCA but first tries are ICA
• Keith thought he ran into an implementation of ICA in Java (fast ICA does integer rather than floating point)
• Todd will look into newer resources for ICA in ImageJ, or others such as Matlab or R.

Better processing through blur and divide

• Designed to equalize a variant background (esp. maps), reduce background as a contributor to PCA and ICA
• Blur each band by some size 50% or more larger than a stroke
• Divide unblurred by blurred

Better processing through correcting registration of captures with filters

• Squashed a bug resulting from ImageJ not adding compression header in tiff when there was no compression
• Software picks squares and computes separation and cross-correlation lateral shift at that location
• From translations at all positions fits a model of magnification and center point
• Some discussion of possible implications of different positions of main bank and aux bank creating minor shadow differences that could be misinterpreted as registration shift

Better capture by identifying fluorescence in IR from visible light

• Ken encountered a dramatic case recently working with modern ink (520nm illumination bright in IR)
• Would be simple to add an IR pass filter and flood with visible light (especially with arc)
• Could slide in an IR pass filter at first, or find a spot on the wheel if UVP filter could be replaced by sum of narrow visible pass filters

Better processing through slave labor

• Nicole P. is doing a 10-week project at RIT (just finished second year at Grinnell, funded by a very selective program)
• Starts May 30, will be invited to June 1 call and R-CHIVE conference

Better scholarship through functional user interface

• The speed issue has been addressed
• Two viewers are supported: Mirador and IIIF Navigator http://jubilees.stmarytx.edu/# viewers

May 4, 2017

Correction for registration loss from UVB and UVP filters (and eventually Wratten).

• Keith has written the software, currently debugging.
• Works by looking at random sample regions in the image and identifying magnification and shift.
• In the long run the tools to perform the correction will be available.
• In the short run Keith will upload corrected UVB and UVP files to the LoadingDock.
• Mike also knows someone who would like better information from a UVP image (David Taylor, Oxford, page from Syriac 30, Old Syriac Gospels)
• Keith will present on this topic at R-CHIVE June 19-20

Viewing environments

• Todd worked on the server and dramatically improved loading times from the IIIF repository.
• No one viewing environment will be best for everyone, but with standards compliance it should be reasonable to support two recommended viewing environments. Todd added a section to the webpage describing the current pros and cons of Mirador (UCLA branch) and IIIF Navigator (http://jubilees.stmarytx.edu/ index.html#viewers).
• Mike has the inside scoop on the development of the UCLA branch of Mirador.
  • They did a lot of work last summer and fall, but subsequently shifted focus to getting images online (about 200,000 images of 74 palimpsests to be done by May 31).
  • After May 31 there will be a series of meetings on functions and tools to set the agenda for development work to resume in the coming budget year, July 1.
  • Susan Marshall and Mike are working with the developers regularly. If Ken and Todd have suggestions for features and improvements it might be best to start through Mike.
  • Documentation should be on their list of things to do, with some pushing by Susan.
• Feedback continues to be welcome

Machine learning

• Some interest at RIT, but best to wait for the end-of-semester chaos to blow over. Revisit in June, mainly at R-CHIVE conference and perhaps leading up.
• Roger discussed the idea with Dave and Anna, both very interested, added to Google doc for June meeting.
• Challenge will be to compete with (or piggy-back on) applications of machine learning with big money
• Mike is reviewing a relevant article for Ira that talks about classes
• Potential for a collaborative doctoral program spanning RIT and Ira

May schedule for this call

• Normal schedule May 11 and June 1, hiatus for May 18 and 25.
• Subsequently options open up over summer (but still best to avoid Fridays and Monday)

Related calls

• Monday May 8 call for Northern Italy projects cooperation set for 3:00-4:00 EDT.
• Schedule for days surrounding June 19-20 will be pushed later this month.
• Roger will work on arranging a time for development office and administrative people to sit down with us before we leave Rochester.

April 27, 2017

Correcting for registration distortion from UVB and UVP Filters

• Great news. Keith is developing a tool in Java that will correct for registration distortion from the filters. No new reference objects or adjustments to the capture procedure are necessary, so we can go back and correct old data.
• It might be ready in about a week. Keith will start with Ambrosiana_F130sup_064v. Roger identified that as a difficult page that could benefit from the new correction.
• The graph paper approach didn’t work because it changes from page to page. The software will handle the correction on page-by-page basis.

Ambrosiana_O39sup (Origen’s Hexapla)

• Keith’s Pseudocolors worked really well
• 001 has reagent that could benefit from Roger’s supervised techniques
• Orientation entered at capture is off in at least some cases

Applying machine learning, neural nets, deep learning, etc. to palimpsests

• Makes a difference whether we are talking about full OCR (characters or even words), vs. pixel by pixel
• Looking for whole characters might be more human-like in utilizing all contextual clues, but even looking at pixels in context would be a way to address the problem that we are capturing more data than humans have time to examine and compare in all possible ways.
• Anna Starynska is a graduate student at RIT working on machine learning applied to OCR
• Dave Messinger’s specialty is graph theory, likely more familiar with tensor flow
• TensorFlow is Google’s neural net toolkit… presumes knowledge of tensors and Python.
• R-CHIVE June 19-20 would be a good time to stand up and see if any computer scientists are interested in taking on as a project
• Google and IBM might be interested in supporting deep learning applied to cultural heritage
  • IBM is interested in applying deep learning to diabetic retinography, also mentioned TensorFlow’s website
  • Could invite someone from Google or IBM to R-CHIVE June 19-20
  • Brent Seales was a research scientist at the Google Cultural Institute, will be at R-CHIVE June 19-20
  • Vinton (Vint) Cerf
  • Mr. B.

Quality time in Rochester

• Should we travel to Rochester May 5-8 as well as June 19-20 to discuss ambitions for Northern Italy projects? Default is no. Original premise was to have our conversation in May and be ready to show off to Mellon rep. in June, but dubious whether we will be ready to demonstrate so soon. Also dubious whether Greg and Dave will have time for us. Might be all we can ask to have a long teleconference on May 8.
• June 19-20 will be the big conference. All on the call plan to at least be there, some will speak.
• r-chive.net
• Also Ira Rabin, Brent Seales, etc.
• All the conversations we are imagining may not fit into two already-busy days (neural net machine learning, northern Italy projects vision articulation, funding strategies).
• Josephine will work on a Google Doc that will outline what meetings should take place so interested parties can plan travel and schedule accordingly.

Mirador, and what it will take before it helps us

• Images take a while to come up (Todd will look into avoiding Amazon S3)
• Needs a written manual with screenshots. A screenshot video may be helpful as a promotional teaser but should not be mistaken for instructional documentation.
• Todd will work on documentation unless someone else is doing so for Sinai Palimpsests or otherwise.
• Mirador and UCLA Mirador are moving targets for feature support and bugs.

Miscellaneous notes

• Files after C73inf with indexes _039 to _042 are dummy place holders for transmissive. They were deprecated from the archive on 4/27/2017.
• The “NoGamma” files could be easily rederived from the flattened but not gamma corrected images using a look-up-table (LUT). This LUT could be incorporated into the SpectralRTI Toolkit, and we could reduce the size of the archive by a third. Ken will keep an eye out for the LUT that Photoshoot uses and Todd will see if it can work in ImageJ.
  • Also note that Jpeg files can be good to have because they are easier to open and transport.
• Ken is going to do some work with law enforcement with respect to forensic applications of advanced imaging. Everyone on the call offered to help.

April 20, 2017

The processing calls had a two-week hiatus due to conflicts, and will resume next Thursday, April 27 at the usual time.

In the meantime, work is continuing on the palimpsests we captured after C73 inf. Keith’s Pseudos and Sharpies have been added to the archive, along with some new files from Roger (http://palimpsest.stmarytx.edu/AmbrosianaArchive/). The archive also includes the semi-annual report I submitted to the NEH (http://palimpsest.stmarytx. edu/AmbrosianaArchive/Reports/ Performance-20170331.html). Images are also being added to the IIIF repository (http://jubilees.stmarytx.edu/uclamirador/). The squashing of bugs is ongoing…

Also, the first publication to utilize our images has already appeared in print. The cover of this book was captured by us.

http://hanut.hebrew-academy.org.il/%D7%9E%D7%95%D7%A6%D7% A8/%d7%93%d7%a7%d7%93%d7%95% d7%a7-%d7%94%d7%a2%d7%91%d7% a8%d7%99%d7%aa-%d7%a9%d7%9c-% d7%aa%d7%a2%d7%aa%d7%99%d7%a7% d7%99-%d7%90%d7%95%d7%a8%d7% 99%d7%92%d7%a0%d7%a1/

April 6, 2017

Lost registration from filters

• Significant issue for UV-Block and UV-Pass filters, less so Wratten filters
• Mike clarified that the noticeable shift he noticed was from one filter to the next in the same sequence
• Still being investigated by Keith is the extent to which a particular filter changes from one sequence to the next in a given camera configuration (i.e., not the intrinsic properties of the filter but its stability and consistency after being spun around).
• One approach would be to stabilize the filters in the wheel (magnets? mechanical slotting?)
• Another approach would be to use reference points in each individual shot to correct the distortion introduced by variation in tilt of the filter
• Will polyester rather than gel filters help?
• Roger tends to avoid UVB and UVP captures because of the registration issue, but Keith will work on fixing registration on a couple of test images for Roger to work on.
• F130sup a good sample of a variety of palimpsests that have not been handled and chemically treated to death
• C73inf 67 and 69 also good candidates because Roger has been working on them (but I think we decided F130sup more likely to be successful)

Keith’s batch processing

• C73inf is complete (Pseudo and Sharpie were not helpful, Ruby was)
• Data for other palimpsests downloaded, not yet processed
• Will be processed with Pseudocolor, Sharpie, Ruby, whatever works

Roger’s focused processing

• Some progress on “Taxo” reading on C73inf page 67 (being uploaded to LoadingDock)
• Ready for some low-hanging fruit… We started with the portions Ceriani did not claim to read, but those portions are not representative of most of the palimpsest. Some “easier” regions could have more scholarly impact (and wow-factor).
• F130sup should be easier, and a good opportunity to experiment with Keith’s filter-distortion-corrected UVP and UVB images
• (I honestly don’t recall why it was important to clarify that O39sup is Origen’s Hexapla and S36sup is Wulfila’s Gothic Bible… both of those are high-value and thoroughly handled.)

Conversation about coordinating for sustainable collaborations

• Mike expects to move Monday’s call to 4:30 pm EST so Roger and Julia Schnieder can join.

March 30, 2017

Keith is still working on using graph paper to correct for distortion from the filters. The problem is they distort inconsistently from one sequence to the next as the filters move about.

The Ruby technique that worked on C73inf may or may not work on other palimpsests. That too is on the list of ways for Keith to not be bored.

Roger has had some interesting results focusing on the hardest portions of C73inf that Ceriani could not read. The next step will be to work on some pages that are less difficult to read but more important to scholars. Page 67 and the relevant coordinates are described on the guide to unread regions: http://palimpsest.stmarytx. edu/AmbrosianaArchive/ GuidesUnreadRegions/ Unreadlines-LatinMoses.html It would be interesting to verify or falsify the reading of Taxo in the specified region. A paleography chart may come in handy: http://palimpsest.stmarytx. edu/AmbrosianaArchive/ GuidesUnreadRegions/ LatinMosesPaleography.html

We’re not returning to Milan this summer, but we haven’t given up. One approach is to resubmit to the NEH. We can make improvements based on new information gathered in January-February. We might be able to secure a better calendar position by planning with don Federico further in advance. This also gives us a window to think more broadly about what we would like to coordinate in northern Italy, and what funding partners may be interested in supporting one or more portions thereof. We can think more broadly than NEH and Mellon, including more regionally focused sources (e.g., https://www.tgci.com/funding- sources/IN/top). Mike and Josephine will try to organize a conference call to begin brainstorming and note sharing about possibilities. We will try to include Todd, Mike, Josephine, Roger, Keith, Ken, Greg, and Julia Schneider (Julia.Schneider.31@nd.edu).

The processing calls will continue next week as usual, except Mike will initiate on Skype because Todd’s Internet connection is in limbo.

March 23, 2017

Following some more work in Oklahoma City, Keith has made progress on using graph paper to correct for registration distortion from UV fluorescence filters. One thought is to disseminate the script for performing the correction rather than the complete set of corrected data. If anyone has a desire to download rather than generate the corrected data, Todd can run the correction on a supercomputer attached to palimpsest.stmarytx.edu. One challenge will be to get the rotation correct, as the folios are either 90° or 270° rotated.

Roger and Todd have been trading notes on targeted processing of lines unreadable in the 1861 edition of Latin Moses. The notes are available here: http://palimpsest.stmarytx. edu/AmbrosianaArchive/ GuidesUnreadRegions/ Unreadlines-LatinMoses.html. Roger’s work has been using blur, divide, and ICA on the transmissive captures. Corrected fluorescence and Ruby images may profitably be added to the mix. So far it looks like a few words are readable beyond what Ceriani could read. More significantly for scholars, it seems likely we will be able to correct some of Ceriani’s claims, which potentially calls all his readings into question. This line of thinking is preliminary and needs verification especially from ruling out the possibility that the letters seen are from the other side. Roger’s images are on the LoadingDock, which is accessible by SFTP but not HTTP. Let me (Todd) know if you need an account.

We can continue with the document based on what Ceriani did not claim to read from Latin Moses. Other paths on the horizon:

• Other passages in Latin Moses which are significant to scholars even though Ceriani offered a reading (one such as it the end of the above document).
• Low-hanging fruit of words that are kinda readable in real life but could be stunningly readable with processing. This would be good for quick presentation of the technology even if not so much discovery. The Arian Commentary on Luke (visually distinct because it is in one column compared to two for Latin Moses) is likely to contain such low-hanging fruit.
• We have other palimpsests to work on. Insight from working on those could bring us back to fresh approaches to C73inf.

Hopefully next week we will know whether another trip to Milan will take place this summer. If so, there may be a case for leaving the Arc of the Firmament in Italy. Don Federico prefers mid-May to mid-June for when such a session would take place.

March 1, 2017

Schedule

• Next two weeks occupied by Oklahoma City
• Next regular call will be March 23, 2:00 pm EDT
• Todd may not be able to make it

Processing

• Overall theme: no magic bullet, different approaches helpful (or not) from one letter to the next
• Keith
  • KTK Ruby (vernacular for Royal Blue fluorescence divided BY transmissive) images are done and published for all 144 pages of C73inf
  • Starting on the data no one has touched yet (A79inf, F130sup, H190inf, O39sup, S36sup)
  • http://palimpsest.stmarytx.edu/AmbrosianaArchive/
• Roger
  • Created 28 images of 18 hard pages (portions unread by Ceriani) using ICA, blur, divide magic
  • sftp://palimpsest.stmarytx.edu/LoadingDock/RLE
  • Waiting for more feedback from Todd about what regions are hopeless with this approach, maybe promising, or very promising
    • So far lower right of C73inf_117 looks good
      • chance of reading beyond Ceriani
      • and perhaps correcting some of what he thought he could read
      • This has big implications because all text is worth checking even if Ceriani offered a reading.
    • So far mostly despair on 112, 129, and 100 with this method
  • More bands on past transmissive sources were helpful
• Todd
  • Working on portions unread by Ceriani and giving feedback to Roger
    • http://palimpsest.stmarytx.edu/AmbrosianaArchive/GuidesUnreadRegions/Unreadlines-LatinMoses.docx
    • Royal blue fluorescence allows reading beyond Ceriani in top right of C73inf_100
    • Being updated with commentary on helpfulness of Roger’s latest magic (above)
  • Started a paleography chart for Latin Moses
    • http://palimpsest.stmarytx.edu/AmbrosianaArchive/GuidesUnreadRegions/LatinMosesPaleography.html
  • UCLA Mirador becoming primary public interface
    • >Some current limitations are being worked around, better long-term viability
    • Latin Moses and Commentary on Luke currently available
    • http://jubilees.stmarytx.edu/uclamirador/

Data Management

• All data captured in Milan is available at http://palimpsest.stmarytx. edu/AmbrosianaArchive/ (and the SFTP equivalent)
• Additional drives in the mail tomorrow morning to Keith and Roger, expected to arrive before leaving for Oklahoma