Fixes
- Fixed RapidOCR missing thin characters such as
<and misreading digits
on pages rasterized at well above--rapidocr-max-side-len, as happens with
--redo-ocr, which rasterizes at 400 dpi. RapidOCR shrank such pages with
interpolation that skips the pixels between its samples, breaking up thin
strokes. OCRmyPDF now shrinks them itself with a Lanczos filter first. On
one scanned assay table, the text layer kept 222 of 229 printed<signs,
up from 163.