Extensions for Piwigo

- Album navigation on the "PDF cover page" tab. A bar of « [ ] Go to next ( ) Left ( ) Right » sits with the cover controls. The arrows step to the neighbouring PDF in the album's own photo order, and Go to next auto-advances there as soon as a cover is set — so a whole album of scanned books can be covered in one pass without going back to the album listing between books. The Left/Right radio chooses which way it advances, for working an album from either end. Remote PDFs registered by S3 Media are included in the queue. - The photo's title is shown on the tab, in the heading and the fieldset legend, so it is clear which book is being covered while stepping through an album. - Historical OCR for books printed in the 1700s-1800s. Early-modern type uses the long s (ſ), which Tesseract's standard eng model does not have in its charset — it substitutes the nearest shape, so Progreſs is read as Progrefs and most of the page is unusable. The settings page can now download frak2021, a model trained on historic Latin-script print — 3.4 MB, CC0 public domain, from UB Mannheim. The file is size- and SHA-256-verified before installation, and written to a temp name and renamed, so an interrupted download cannot leave a half-written model in place. - OCR language / model setting. Chooses the model passed to tesseract -l, validated against the models actually installed. Defaults to eng, so existing installations behave exactly as before. Once the historical model is present, combined entries (eng+frak2021, fra+frak2021) are offered too. Measured on a 1724 title page: eng read 11/14 ordinary words but 0/5 long-s words; frak2021 read 5/5 long-s but only 6/14 words; eng+frak2021 kept 11/14 words and recovered the long s. - Even out the lighting before OCR setting (on by default). Scanned books are rarely evenly lit, and Tesseract's global thresholding drags shadowed text under with the shadow — on a real 1848 scan the first word of a title line vanished. An adaptive local threshold judges each pixel against its own neighbourhood and restores it. - Convert historical characters to modern equivalents setting (on by default). Folds the long s and historical ligatures to modern letters so the text is findable by Piwigo's search. Deliberately maps only ſ and never f, which would corrupt every genuine f in the text. - Enlarge the image before OCR setting (on by default). Covers are rasterised at 150 dpi, half the 300 dpi Tesseract documents as its floor. The image is upscaled and deskewed into a throwaway file before recognition, then deleted. - OCR timeout (seconds) setting (default 60). The OCR call previously had no timeout at all, so a pathological image tied up the request until PHP's max_execution_time. - Fixed: thumbnails no longer show the previous cover after a cover is changed on a remote PDF. The cache-buster only fired for representative paths, but a remote PDF's cover overwrites the poster original — so its URL never matched and the browser kept serving the cached image. - Fixed: the cover tab no longer renders a bare "› Edit photo ›" heading over an empty form and a broken thumbnail when a PDF fails a guard. - Fixed: the settings page no longer fatals with "Cannot access offset of type string on string". admin/settings.php runs in Piwigo's global scope, where a local named $lang overwrote core's translation array. - Fixed: selecting a downloaded model no longer hides the system ones. TESSDATA_PREFIX replaces Tesseract's search path rather than extending it, so pointing it at the plugin's directory made -l eng fail. - Fixed: README and the marketplace description no longer claim the plugin "stores no configuration" — it has had settings since 16.7.0 — and both now list Tesseract as the requirement for the optional OCR button.