Project 2 · App + analysis workshop
MIS, the smart magnifier for the sofer
Magnifier Intelligent for Sofer (formerly SoferOCR). A help for checking STaM scrolls: mezuzot first, then tefillin, megillot and Sifrei Torah.
Prototype
This is not OCR
A classic OCR, trained on printed Hebrew, cannot read a sofer’s Ashurit script. But the text of a mezuzah is known in advance: 713 letters, 170 words, in a fixed order.
So the question is not “what is this letter?” but “does this sequence of ink shapes match the expected text?”. It is an alignment problem, far more reliable: even if each isolated letter is poorly recognised, the context of hundreds of neighbouring letters makes it possible to recover the exact text.
MIS never declares a scroll kosher. The app reports measurable differences from the reference text and shape anomalies. The halachic judgement always belongs to the magiah.
How it works
- Photo or scanof the scroll, flat or with a phone
- Cleaningscaling and binarisation of the ink
- Linesdetection of each line of text
- Letterssegmentation of the letters, right to left
- Recognitionprobabilities over the 27 signs
- Alignmentcomparison with the reference text
- Reportmissing, extra, doubtful or touching letter
Where are we?
Prototype
- Complete analysis pipeline in a PC workshop
- Tests on synthetic scrolls
- Learning the script (Beit Yosef) on certified mezuzot
- Review of the reference text by a sofer
- Mobile app with the camera
- Tefillin, then megillot and Sifrei Torah
- Eventually: a dedicated scanner
We are looking for
Volunteers, at your own pace, from home. Every helping hand moves the project forward.
Does this project speak to you?
Developer, rabbi, sofer, translator, tester or simply motivated: write to us, we will answer.