hocr-pdf : Possible calculation issue

[whikloj]

I could be wrong, but in reading this calculation which you use for adjusting the height of text it seems like box[0] is left and box[2] is right from the bbox coordinates. Additionally, the linebox[0] would also be left.

I changed it to this based on my reading of the hOCR spec for bbox

But in case I misunderstood your intention, I thought I'd open this issue.

[zuphilip]

What will this change in the outputted PDF? CC @jbreiden @kba

[whikloj]

So if I am understanding correctly.

Your calculation is

b = polyval(baseline, (box[0] + box[2]) / 2 - linebox[0]) + linebox[3]

which based on the HOCR spec is

b = polyval(baseline, (bbox-left + bbox-right) / 2 - linebox-left) + linebox-bottom

which you then use here

text.setTextOrigin(box[0] * 72 / dpi, height - b * 72 / dpi)

So this calculation is using Left and Right to calculate an average then subtracting that from height. It seems to me the you'd want to get the average of Top and Bottom (ie. box[1] and box[3]).

So to my mind this makes more sense

b = polyval(baseline, (box[1] + box[3]) / 2 - linebox[1]) + linebox[3]

or

b = polyval(baseline, (bbox-top + bbox-bottom) / 2 - linebox-top) + linebox-bottom

The difference would be seem more obviously in longer words where the difference between Left and Right would be larger. But (as I said before) perhaps I am not understanding what you are trying to accomplish with this calculation.

[jbreiden]

I don't actively use this program any more, so have not been paying attention. Test with descenders like 'yyy', without descenders like 'xxx' and mixed like 'xxxyyy'.

[nsshah14]

@zuphilip @jbreiden Can you elaborate on this calculation. I am not getting the proper selection for my text.

Here the selection goes to the bottom of the text.