09-05-2020, 06:37 AM
Hi Anton,
Yes, agreed that the positional analysis is cumbersome. One could define (0,0) as the position of the top left glyph on the folio. Each glyph on the folio would have a bounding box, in millimetres or pixels, whatever is more convenient. And each glyph would be offset by (x,y) from the top left glyph. Similarly, each word would have a bounding box, comprised of the envelope (probably convex hull) of its glyphs' bounding boxes. And, the various blocks would also have bounding boxes, or polygons. With this data one could ask how densely packed words are in a given block, and one could make stats on which glyphs appear in which folio locations or word blocks. I feel like I've seen a similar proposal (to include absolute positional info in a transcription) somewhere years ago, perhaps by Rene?
Ideally an image analysis library could do all this work - a folio by folio positional analysis of glyphs, words, blocks - but it seems computationally challenging. Perhaps one of the adversarial neural nets would do a good job? Or, find a willing student to do it by hand, LOL.
Julian
Yes, agreed that the positional analysis is cumbersome. One could define (0,0) as the position of the top left glyph on the folio. Each glyph on the folio would have a bounding box, in millimetres or pixels, whatever is more convenient. And each glyph would be offset by (x,y) from the top left glyph. Similarly, each word would have a bounding box, comprised of the envelope (probably convex hull) of its glyphs' bounding boxes. And, the various blocks would also have bounding boxes, or polygons. With this data one could ask how densely packed words are in a given block, and one could make stats on which glyphs appear in which folio locations or word blocks. I feel like I've seen a similar proposal (to include absolute positional info in a transcription) somewhere years ago, perhaps by Rene?
Ideally an image analysis library could do all this work - a folio by folio positional analysis of glyphs, words, blocks - but it seems computationally challenging. Perhaps one of the adversarial neural nets would do a good job? Or, find a willing student to do it by hand, LOL.
Julian