END OF THE PROJECT

My part of the project is available here (prerequisites, steps and final code version) plus new glyphs dataset self-elaborated from ‘Matricula de Tributos’ codex

Finally, the code was deployed also in UOregon’s server, as AztecGlyphRecognition URL

And a functional mirror-URL is provided in Heroku for Aztec Glyphs Recognition v.2

Further steps

As a result of our summer work, these are the future steps that might be done:

  • Normalized all future image filenames with same pattern.

  • Do not use forbidden characters in filenames such as “&, !, “, ‘, $, ?” and so on.

  • Peripherical material could show abnormal results due to the low glyph variability (in some cases no more than 2 o 3 examples in dataset)

  • Transparent background is converted to white by the algorythm. So white background should be the best option due to a RGB (0 to 255) white is (255,255,255) value and black is (0,0,0). So any peripherical material or background color in user’s image will have same cosine distance to glyph dataset in full white background

  • Mobilenet algorythm using keras was acceptable for this early version. But due to comparson of full visual pixel matrix, another algorythms should be more accurated. Just as an example, another Red Hen Lab GSoC 2022 project by Rishab Mudliar could fit with this task.

  • Next codex to incorporate due to high percentil similarity will be Osuna Codex, Telleriano-Remensis, Vaticanus A, Matrícula de Huexotzinco, Tovar or Aubin, just as examples