Scantpaper

A GUI to produce PDFs or DjVus from scanned documents.

Scantpaper is a Linux application (it needs GTK4, SANE, and a Python 3 interpreter, all of which are available on other Unix-like systems like MacOS or BSD as well). It is the Python rewrite (v3) of the popular gscan2pdf.

• Scan single- or double-sided, with automatic interleaving
• OCR with tesseract for searchable PDF/A, without the need for temporary files
• Edit with crop, rotate, threshold, unsharp mask, and unpaper clean-up
• Save as PDF, DjVu, TIFF, PS, TXT, hOCR, or image files
• Recover crashed sessions and restore them on the next start

Screenshot: Main page

Table of Contents

• Quick Start
• Description
• Command-line Options
• Diagnostics
• Configuration
• Dependencies
• Download, Installation & Removal
• Support
• Reporting Bugs
• Translations
• FAQs
• Known Limitations
• See Also
• History
• Author
• Thanks To
• Contributing
• Donate
• License

Quick Start

Install scantpaper and its dependencies (see Download, Installation & Removal), then:

1. Start the application with scantpaper (or python3 -m scantpaper.app from a source checkout, with PYTHONPATH=src set). Add --debug|info|warn|error|fatal to enable logging at the required level.
2. Scan one or several pages with File → Scan. Activating Scan while scan options are still being applied starts the scan when the apply finishes, so a second click is not needed.
3. Select the pages and create a PDF with File → Save.
4. To make the saved PDF searchable, enable OCR in the scan window or run Tools → OCR before saving.

Description

scantpaper provides a GUI for scanning, editing, and saving documents as PDF, DjVu, TIFF, PS, TXT, hOCR, SDB (scantpaper session), or image files (PNG, JPEG, PNM, GIF), and can prepend or append to an existing PDF. It supports batch scanning, metadata, OCR, and various editing tools.

How it works

Scans are acquired with SANE and held in a session database while you edit them. When saving, PDFs are produced with img2pdf and OCR’d with ocrmypdf (which produces PDF/A out of the box); DjVu export uses djvulibre-bin, TIFF export uses libtiff, and images are written with ImageMagick.

┌─────────┐   ┌─────────────────┐   ┌──────────────┐   ┌──────────────────┐
│  SANE   │   │ SQLite session  │   │ edit tools / │   │ img2pdf /        │──▶ PDF (PDF/A)
│ scanner │──▶│ (pages in temp  │──▶│ OCR          │──▶│ ocrmypdf         │
│         │   │  directory)     │   │ (tesseract)  │   ├──────────────────┤──▶ DjVu
└─────────┘   └─────────────────┘   └──────────────┘   │ djvulibre-bin    │
                                                       ├──────────────────┤──▶ TIFF
                                                       │ libtiff          │
                                                       ├──────────────────┤──▶ PNG, JPEG, PNM, GIF
                                                       │ imagemagick      │
                                                       └──────────────────┘

Page Numbering

Page numbers are always consecutive (1, 2, 3, …). Deleting a page renumbers the remainder automatically, and editing a page’s number in the document table moves that page to the corresponding position.

Scan Flow

• Single-sided: Each scan is appended at the end.

• Double-sided: Scan all facing pages first (front 1, front 2, …, front n); they are appended in order. When you flip the stack the ADF feeds them in reverse (back of page n, then back of page n-1, …, back of page 1). Each reverse page is inserted immediately after its matching front page, producing a fully interleaved result: front 1, back 1, front 2, back 2, …, front n, back n.

  scan pass 1 (fronts):     front 1, front 2, ..., front n
  flip stack
  scan pass 2 (backs):      back n, back n-1, ..., back 1
  interleaved result:       front 1, back 1, front 2, back 2, ..., front n, back n

• Extended mode (insert before page N): Each new scan is inserted before the selected page, advancing the insertion point for the next scan.

ADF / duplex scanning

Automatic document feeder (ADF) and duplex scans now import every side of the document. Some Brother scanners (e.g. the DS-740D) prefetch the reverse side of a sheet as soon as the front side finishes reading; cancelling the scan session between pages discarded that buffered side, so only the first page of a duplex job was imported. Scantpaper no longer cancels the session between pages when scanning from a feeder, which preserves the prefetched side. The session is still cancelled at the end of the batch, when the requested page count is reached, on error, or when you cancel the scan.

For multi-page flatbed batches the behaviour follows gscan2pdf semantics: the Force new scan job between pages preference (enabled via Edit → Preferences, only available when Allow batch scanning from flatbed is enabled) controls whether the session is cancelled between flatbed pages; the session is always cancelled at the end of the batch. Switching to a flatbed source starts # Pages at one page. To scan an unbounded flatbed batch, choose All explicitly; the dialog then warns that scanning continues until you cancel. Cancelling stops the batch after the page already in progress. If the scanner does not respond to cancel, scantpaper reports the failure and rebuilds the device session for the next scan; a device still held by the operating system is reported as busy rather than leaving the application unable to scan.

When scanning a book destructively (cutting off the binding and aligning every untorn outer edge to the same rim), enable Alternate rotation every 2nd page in the Postprocessing tab. Odd pages then use the configured rotation angle, even pages use that angle plus 180 degrees, so the flipped back of each sheet comes out upright too. The toggle is only visible for flatbed scanning (including one-page-at-a-time manual scans) when batch scanning from the flatbed is allowed; rotation still runs before OCR. In manual (single-page) mode the alternating parity persists across consecutive scans until you explicitly turn the toggle off.

Hiding unwanted devices

SANE backends sometimes list a webcam or another non-scanner as a device. Select it in the Device list and click Blacklist this device to hide it; the pattern is stored in the Device blacklist field in Edit → Preferences. That field also accepts a regular expression, matched case-insensitively against the label shown in the list and the device’s backend name, so patterns such as Camera or v4l:.* work. If a pattern hides every device, scantpaper reports that no devices were found rather than offering an empty list.

Main Features

• Scan: Options for device, page count, source document, side to scan, and device-dependent options (page size, mode, resolution, batch-scan, etc.). Optionally rotate (including alternating by page parity for book flatbed scans), clean up with unpaper, or run a user-defined tool on each page on scan, and OCR each page. When applying a scan-options profile, the scanner driver may revert an option (some drivers couple or alias options, e.g. page size and scan area); such an option is then set at most twice and dropped for the rest of that apply, so the scan still proceeds instead of ending in a reload-recursion error. If the scan options request 1-bit output - a 1-bit depth, or a bilevel scan mode such as Lineart - the delivered pages are checked against that request. Bilevel transfers always arrive as 8-bit grayscale (the SANE front-end expands them, even when the driver sends 1-bit data, for example when a hardware compression option is active), so scantpaper converts each such page to black-and-white at scan time, using the colour-aware Threshold semantics and value, shows a notice in the progress bar, and logs a warning - so “Automatic” PDF compression encodes the pages as Group 4 or JBIG2 rather than JPEG. Scan-area and page-size fields accept fractional values typed with your locale’s decimal separator (e.g. 115,2 mm in a German locale), show them without trailing zeros (174 mm, not 174,00), and step by whole units with the arrows while keeping the typed fraction. Only separators your locale actually uses are accepted when typing numbers; a period in a comma locale is rejected rather than silently misinterpreted. The same fractional entry applies to the page-properties resolution fields and to the blank/dark-page and unpaper white/black thresholds, so e.g. 300 dpi and 0,9 are shown exactly as typed.
• Save: Save selected/all pages in multiple formats. Supports metadata. The Title, Author, Subject, and Keywords fields offer autocompletion, suggesting values from imported documents and values you have entered before. When saving as PDF, the progress bar tracks each page as it is written and reports the PDF conversion step. Imported images are stored in a compact format (JPEG for scanned pages, the original bytes for imported JPEG/PNG files, lossless PNG for bilevel, transparent, or posterised grayscale pages). When saving as PDF, the default Automatic compression picks an encoding per page: lossless CCITT Group 4 for bilevel pages, lossless Flate for images with transparency or posterised (few-level) grayscale, and JPEG for other continuous-tone colour or grayscale pages at the configured Quality. The Compression menu also offers an explicit codec for every page: G4, Flate, JPEG and JPEG 2000, plus JBIG2 when the optional jbig2 encoder is installed. JPEG 2000 usually produces smaller files than JPEG for scanned pages but is slower to encode and less widely supported by viewers; JBIG2 typically halves the size of a bilevel page compared with Group 4. scantpaper always uses JBIG2’s lossless mode (see JBIG2 and the Xerox character-substitution bug), so the saved bitmap is unchanged; the exact output size depends on the installed encoder version, and like G4 it thresholds grayscale and colour pages to bilevel first. The Bilevel compression preference in Edit → Preferences (Group 4 by default, JBIG2 when available) tells Automatic how to encode bilevel pages, so a user who never changes the preference keeps Group 4 output. The Quality spin-button steers both JPEG and JPEG 2000, and lower values shrink those streams. Unmodified imported JPEG pages are embedded byte-for-byte, while pages that scantpaper generated by scanning or editing are re-encoded at the configured quality, so lowering Quality shrinks the PDF. A lossless encoding (Automatic’s Group 4 or Flate handling, or the explicit Flate option) is preserved through the PDF/A conversion, so the page image is not silently transcoded to JPEG, on every supported OCRmyPDF version. Saving to a file that already exists asks for confirmation first: declining leaves the existing file untouched and keeps the save dialog open, and a save that fails reports the error instead of silently doing nothing. When saving as PDF, pages are processed one at a time and page images are decoded on demand, so saving a document with many high-resolution pages no longer keeps every decoded page in memory and no longer exhausts the memory of the process.
• Import hOCR: Apply a saved hOCR file to the selected pages or all pages. Pages are matched in order, the page box must match the target page’s pixel size, and only the text layer is replaced; images and annotations are left untouched. This is useful for restoring corrected OCR after rebuilding a session from the original images.
• Thumbnails: Page thumbnails are generated with a decimate-then-LANCZOS downscale for a sharp preview in the thumbnail panel without slowing down bulk imports, and bilevel or palette pages (common in scanned/OCR PDFs) are anti-aliased so their previews stay legible.
• Lists: Typing with a list widget focused (the thumbnail panel, or a picker such as the File → Open crashed session dialog) no longer opens GTK’s interactive-search overlay, which showed nothing because no list configures a search column. It also removes the gtk_css_node_insert_after Gtk-CRITICAL that GTK 4.24 printed in its place on the first such keystroke.
• Email as PDF: Attach pages as PDF to a blank email (requires xdg-email).
• Print: Print selected/all pages.
• Dialogs: Every dialog insets its content and its action buttons from the window edge, spaces its rows apart and spaces the controls within a row apart, so no control touches the window border and a label no longer sits flush against the field beside it. Action buttons are also separated from each other; a message dialog keeps its joined button bar. In the scan, Clean up and Preferences dialogs, rows keep the height their content needs when the window is taller than the page: the surplus appears as empty space below the last row instead of stretching a frame, checkbox or label-and-control row. A message dialog keeps its “Don’t show this message again” control in a fixed row below the scrolling list, gives the spare height to the list, and shows each message as selectable text. A message without a page number leaves its Page cell empty instead of showing “None”. Escape and the window close control close it with a response, so the caller returns and the dialog does not trip the GTK4 IBus module’s focus-out crash.

JBIG2 and the Xerox character-substitution bug

JBIG2 is a format for bilevel images (ITU-T T.88 / ISO/IEC 14492), not a single compression mode. It defines lossless and lossy coding. In the lossy mode an encoder can replace similar-looking shapes with one shared symbol and discard the difference; if that matching is done badly, a 6 can decode as an 8. That is what happened in 2013 on some Xerox WorkCentre and ColorQube devices, whose firmware used lossy pattern matching and substitution. The fault was in the encoder, not in JBIG2 as a format; David Kriesel, who reported the bug, concluded that “the error cause is not JBIG2 itself”.

scantpaper always uses the lossless path. It converts only bilevel pages, one image at a time, with no shared symbol dictionary and no lossy substitution, and it never enables OCRmyPDF’s lossy JBIG2 option. The JBIG2 step therefore reproduces the bilevel bitmap exactly. OCRmyPDF 17 removed its lossy JBIG2 support for the same reason and now encodes each image independently and losslessly.

“Lossless” describes the compression step, not the whole scan. If you choose JBIG2 for a grayscale or colour page, scantpaper thresholds it to bilevel first, exactly as it does for Group 4, and that thresholding is where detail is lost. No compression setting can repair a scan that was already wrong.

PDF/A does not require lossless compression. ISO 19005 permits lossy compression, and PDF/A allows JBIG2 in both its lossless and lossy forms, so a PDF/A stamp is not evidence that a scan used the lossless mode. scantpaper’s PDF/A output is lossless because scantpaper asks for the lossless encoder path, not because PDF/A forces it. Some archival guidelines are stricter still: Germany’s BSI TR-03138 for replacement scanning forbids pattern matching and substitution, and also soft pattern matching; together those rule out JBIG2’s symbol-coding modes, including the lossy pattern-matching-and-substitution method behind the Xerox bug. That is a scanning policy, not a PDF/A rule, so check the recipient’s requirements if you scan for one.

Further reading:

• JBIG2 — Wikipedia, especially “Character substitution errors in scanned documents”.
• Xerox § Character substitution bug — Wikipedia.
• Xerox scanners/photocopiers randomly alter numbers in scanned documents — David Kriesel’s original report.
• ITU-T Recommendation T.88 — the JBIG2 standard.
• ISO 19005-2:2011 (PDF/A-2) — its introduction notes that a conforming file may use lossy compression.
• PDF Association, PDF/A FAQ — confirms that compression, including lossy compression, is permitted.
• OCRmyPDF: Installing the JBIG2 encoder — states that OCRmyPDF uses lossless JBIG2.

Edit Menu

• Undo, Redo: Undo or redo the last action. An operation applied to several pages at once (for example rotating or cropping a selection) is undone or redone as a single step, reverting every affected page together. Text and annotation corrections are undone and redone the same way; undoing an edit also restores the page’s previous saved state. Clearing the document with New File is itself an undo step: undo right after it restores the cleared pages, and undoing a later edit never resurrects pages that were cleared before it.
• Cut, Copy, Paste: Cut, copy, or paste selected pages.
• Delete: Remove selected pages.
• Select: Select all, odd, even, inverted, blank, dark, or modified pages, or pages without (up-to-date) OCR.
• Properties: Edit image metadata.
• Preferences: Configure default behaviors and frontends.

View Menu

• Pan, Select, Select & Pan tools, for both the image and the text-layer or annotation canvas:
  • Pan: Use the left mouse button to drag the image or canvas to pan the view. The middle mouse button does nothing.
  • Select: Use the left mouse button to select a rectangular box. The middle mouse button does nothing.
  • Select & Pan: Use the left mouse button to select a rectangular box, and the middle mouse button to drag the image or canvas to pan the view.
  • In all of the above, the mouse wheel zooms in or out.
  • When no page is displayed in the image view, pressing or dragging does nothing.
• Clear image selection: Remove the selection rectangle from the image and layer panes. The same action is also on the toolbar next to the selection tools and available from the keyboard with Escape.
• Zoom: 100%, fit to window, in, and out.
• Rendering: pages are shown at high quality whenever they are static, so the full-page view is easy to assess; a faster rendering is used only while you are panning or zooming, for smooth interaction.
• Rotate: 90° clockwise, 180°, and 90° anticlockwise.
• Layer editing: choose Edit text layer or Edit annotations to work on that layer. In the tabbed view the selected tab decides which layer is shown, and its control bar appears below the image only while that tab is open (on the Image tab neither layer is selected and no control bar is shown). In a split view the chosen layer is displayed beside the image, so the annotation layer can be corrected next to the scan as well as the text layer. The layer you were last editing is remembered when you restart. The control bar has a fixed height: it appears or disappears with its layer immediately, and cannot be resized by dragging. Click a slice to focus it for correction; with the Select or Select & Pan tool active, press or drag anywhere else in the pane to start a new selection rectangle, while the Pan tool pans instead. Focusing a slice zooms so its line is drawn at a consistent, comfortable reading size rather than a fixed fraction of the pane, so short and long words from the same line look alike and a long word still fits with a margin around it. The zoom button beside the navigation controls opens a Focus zoom slider (50-200%, default 100%) that adjusts that size while you work. Reset returns it to 100%, and the chosen level is remembered across restarts and applies to both the text and annotation layers.

Keyboard Shortcuts

Tools

• Threshold: Binarize images. The value is an ink-strength cutoff: pixels that differ from the paper colour by more than the given percentage are rendered black, so coloured text and annotations (stamps, highlighters) on white paper are kept. The default is 20; values saved by older versions are migrated automatically.
• Brightness / Contrast: Adjust brightness and contrast.
• Negate: Invert colours.
• Unsharp mask: Sharpen images.
• Crop: Crop selected pages. The selection rectangle stays visible when you move between pages, so you can preview where the crop will land on each page before applying it.
• Clean up: Use unpaper to clean up scans.
• Split: Split pages vertically or horizontally.
• OCR: Use tesseract to create a text layer for the selected pages. The text layer is embedded into saved PDFs, making them searchable, and can be viewed and edited in the text layer window. Deleting a slice in that window removes the corresponding text from the saved PDF’s text layer, and correcting one replaces the text that is saved and drawn; emptying a slice deletes it. Slices added manually are preserved in the saved text layer too, even when they sit inside an existing OCR paragraph. Accepting a correction moves on to the next slice in the current sort order, so a confidence-sorted pass walks from the least to the most confident text without re-selecting each one. To add a slice, first draw a rectangle - on the image or directly in the text layer pane - then type the text and press the + control; the Add control stays disabled, with an explanatory tooltip, until a rectangle has been drawn or selected. Drawing that first rectangle works on a page that has not been OCRed yet, so the text layer can be started from scratch. A focused slice can be rotated 90° clockwise or anticlockwise to author vertical text (e.g. rotated index columns) whose glyphs tesseract read wrongly; slices added while a slice is rotated inherit that rotation. A slice added without an inherited rotation is oriented automatically: a tall, narrow selection becomes vertical text and a wide, short one stays upright, so a drawn box for a vertical column renders legibly instead of as shrunken horizontal text. Rotated slices are embedded at their angle in saved PDFs. Upright and rotated slices can be mixed on one page: each word is saved at its own size and angle, so a search highlight covers the word it found rather than a larger block around it. Re-importing a PDF cannot recover that rotation, because the imported text layer carries positions but not text direction. Page pixels are fed to tesseract in memory and the stored image is reused, so OCR is faster than in previous versions.
• User-defined: Run user-defined commands.

User-defined Tool Variables

• %i - input filename
• %o - output filename
• %r - resolution

By default a user-defined tool is run once per selected page, with %i (and %o/%r where used) substituted for that page. Selecting “Process selected pages at once” in the user-defined tools dialog instead passes every selected page to a single invocation: %i expands to all the working files, space-separated, so a tool that edits files in place (such as mogrify -negate %i) can process the whole selection in one go. Only the pages whose file actually changed are replaced, and the whole batch is a single undo step, so one undo restores every page. %o and %r are not available in batch mode; a tool that uses them must be run one page at a time.

OCR Languages and Tesseract Options

Tesseract can recognise several languages at once. In the scan window and the Tools → OCR dialog, pick one Language to recognise (the primary language) and tick any Additional languages in the popover. The resulting specification is stored in the ocr language setting as a +-joined list, for example deu+eng. The primary language is listed first, so put the main language of the page first; tesseract prioritises the first language, and adding more languages makes recognition slower and can reduce accuracy, so select only the languages you need.

Next to the language, two selectors control how tesseract runs:

• Page segmentation mode tells tesseract how to lay out the page, from Automatic (the default, suitable for most pages) through Single column, Single block, Single line, Single word, Single character, Sparse text, and Raw line.
• OCR engine mode chooses the engine: Default (the engine’s own choice), LSTM only, Legacy, or Combined. Legacy needs language data that includes the legacy engine; if it is missing, OCR does not start and the error names the engine mode.

Both are stored and restored with the other OCR options. Do not use tessedit_pageseg_mode or tessedit_ocr_engine_mode in the advanced field instead: those variables are set by these controls, so the field rejects them.

The Advanced Tesseract engine parameters expander accepts one name=value engine parameter per line, for example:

user_words_file=/home/me/words.txt
tessedit_char_whitelist=0123456789

These are the values of tesseract’s -c option, passed straight to the engine, and cover options that have no dedicated control, such as a user word list (user_words_file) or a user pattern list (user_patterns_file). A leading -c is optional, so -c user_words_file=/path works too.

Press Tab to leave the field for the next control, so OK (or Scan) stays reachable without the mouse. Ctrl+Tab and Ctrl+Shift+Tab move focus forward and backward too. A literal tab character is not needed.

Do not paste tesseract command-line options such as --user-words: the field expects engine parameters, and the error names the parameter to use instead:

Lines beginning with # are comments; blank lines are ignored; a value may itself contain =. Paths are not expanded, so use absolute paths. A user_words_file or user_patterns_file that does not exist or cannot be read is reported before OCR starts, and the file paths that will be applied are written to the log. Invalid or unknown options are reported before any page is modified.

The Available parameters button below the field opens a browser of the engine parameters that the installed Tesseract reports, with each parameter’s name, default value, and description. Nothing is bundled with the application, so the list follows the installed engine. Type in the filter to narrow the list by name or description; matching is case-insensitive, and a filter that matches nothing says so. Select a parameter and press Enter or Insert to write name= at the cursor in the advanced field and return to the field with the cursor after the =, ready for the value. If there is already text on the current line, the new parameter starts on its own line. tessedit_pageseg_mode and tessedit_ocr_engine_mode stay visible but are annotated with the control that sets them and cannot be inserted. If the list cannot be read, the browser says so and OCR still runs with the parameters already in the field.

The tessdata_best language models give more accurate OCR but are larger and notably slower, and they only support the LSTM engine. Point the Tessdata directory control (see below) at their directory to use them; the TESSDATA_PREFIX environment variable and replacing the system models also work.

Tessdata directory

The Tessdata directory control at the top of the OCR options chooses the directory that holds the tesseract language data (the .traineddata files). Leave it empty for the automatic directory, shown in the entry as Default (system tessdata); Browse picks a folder and Default clears the choice.

The directory is resolved in this order:

1. the chosen Tessdata directory;
2. the TESSDATA_PREFIX environment variable, if it is set;
3. the system tessdata directory.

A chosen directory must directly contain at least one .traineddata file. This also applies to TESSDATA_PREFIX: point it at the directory that contains the files, not at a parent such as /usr/share. A broad value makes tesseract search the whole tree and hang, so it is rejected with an error instead of being used. Nothing falls back to another directory: a wrong path is reported before OCR starts.

Changing the directory rebuilds the language list from it. If the current language is not present in the new directory it stays selected and is marked (not installed), and OCR refuses to start until a language from that directory is chosen.

The selected directory is stored with the other OCR options and restored on the next start. Tesseract 3’s m_data_sub_dir parameter no longer exists in Tesseract 4 and later, so it is rejected like any other unknown engine parameter.

Command-line Options

scantpaper supports the following options:

• --device <device> [...] Specifies the device(s) to use, instead of getting the list from the SANE API. Useful for remote scanners. May be repeated, or given multiple space-separated devices.

• --help
  Displays help and exits.

• --log=<log-file>
  Specifies a file to store logging messages. On exit, the log is compressed to <log-file>.xz.

• --debug, --info, --warn, --error, --fatal
  Defines the log level. Defaults to --debug if a log file is specified, otherwise --warn.

• --import=<PDF|DjVu|images>
  Imports the specified file(s). For multi-page documents, a window is displayed to select required pages.

• --import-all=<PDF|DjVu|images>
  Imports all pages of the specified file(s).

• --locale=<directory> Sets the directory containing translated messages. See Translations.

• --version
  Displays the program version and exits.

Example output

$ scantpaper --version
scantpaper X.Y.Z

(Replace X.Y.Z with your installed version.)

$ scantpaper --help
usage: scantpaper [-h] [--device DEVICE [DEVICE ...]]
                  [--import IMPORT_FILES [IMPORT_FILES ...]]
                  [--import-all IMPORT_ALL [IMPORT_ALL ...]] [--locale LOCALE]
                  [--log LOG] [--version] [--debug] [--info] [--warn]
                  [--error] [--fatal]

A GUI to produce PDFs or DjVus from scanned documents

options:
  -h, --help            show this help message and exit
  --device DEVICE [DEVICE ...]
  --import IMPORT_FILES [IMPORT_FILES ...]
  --import-all IMPORT_ALL [IMPORT_ALL ...]
  --locale LOCALE
  --log LOG
  --version             show program's version number and exit
  --debug
  --info
  --warn
  --error
  --fatal

Please see /usr/share/doc/C/scantpaper/documentation.html for more detail

Examples

# Import every page of a PDF, letting you edit before saving
scantpaper --import-all ~/scans/document.pdf

# Import a PDF, choosing the pages to import in a dialog
scantpaper --import ~/scans/document.pdf

# Use a remote scanner
scantpaper --device "net:scanner.example.com:6566"

Scanning is handled with SANE. PDF conversion uses img2pdf and ocrmypdf. TIFF export uses libtiff.

Saved PDFs are marked as created by scantpaper: the document info Creator field and the XMP creator tool begin with scantpaper v<version>, followed by the OCR toolchain used, e.g. scantpaper v3.0.16 / OCRmyPDF 16.13.0 / Tesseract OCR-hOCR 5.5.0. The Producer field continues to name the library that wrote the file.

Diagnostics

To diagnose errors, start scantpaper from the command line with logging enabled:

PYTHONPATH=src python3 -m scantpaper.app --debug

Configuration

scantpaper creates a config file at ~/.config/scantpaperrc. The directory can be changed by setting $XDG_CONFIG_HOME. Preferences are usually set via Edit → Preferences.

If the config file cannot be read as JSON (for example after a partial write or a hand edit), scantpaper rescues as many recognised settings as it can and resets the rest to their defaults. The unreadable original is kept as a backup at ~/.config/scantpaperrc.old, and the start-up message tells you which settings were kept. Wrongly typed values (e.g. a string where a number is expected) are corrected when possible; each correction is written to the log rather than shown in the start-up message. Only values that cannot be corrected are reported. The original file is never silently overwritten: saving on exit waits until the message has been acknowledged.

Scan options saved in the configuration file by gscan2pdf 2.x (whose default-scan-options block has no frontend list) are recognised and applied to the scan dialog on startup, and are kept when the file is written back.

Sessions

An unsaved session’s data (pages, edits, OCR, annotations) is stored in an SQLite database in a temporary directory named scantpaper-????????, created under $TMPDIR (or /tmp) by default. The database is a sibling of that directory: /tmp/scantpaper-????????.sdb, not a file inside it. You can change the temporary location in Edit → Preferences.

On a normal exit the temporary directory, the database and its sidecars are removed. If scantpaper crashes, they survive. The next start opens a Crashed sessions picker, and File → Open crashed session opens the same picker later. A snapshot that another instance is using is not listed.

The picker offers Restore, Discard and Cancel:

• Restore copies the snapshot into a fresh temporary session and opens that copy. The snapshot and its directory are removed only after the restored pages are available, so a failed restore leaves the snapshot intact. The restored work is an ordinary unsaved temporary document: it is removed on a normal exit unless you save it.
• Discard asks for confirmation and then removes the selected snapshots, their sidecars and their session directories.
• Cancel leaves everything unchanged.

A snapshot whose latest action removed every page is still listed, because its undo history may contain recoverable pages. Restoring it opens an empty document; discarding it removes it.

To remove a leftover snapshot manually, quit scantpaper first and delete both the database and its directory:

rm -rf /tmp/scantpaper-???????? /tmp/scantpaper-????????.sdb*

A session saved as an .sdb file (File → Save As) can be reopened with File → Open. The opened file itself becomes the working session: edits are written to it as you work, so they persist even if you quit without saving the session again, and quitting does not warn about unsaved pages. Saving over an existing .sdb file asks for confirmation before the old file is replaced; the replacement is atomic, so a failed save leaves the previous file intact. Because the opened file is the source of truth, an in-place session is not offered for restoration as a crashed session after a crash.

To reuse corrected OCR when rebuilding a session from the original images, save the old session’s pages as hOCR, import the images into a new session in the same page order, and choose File → Import hOCR… for the selected pages or all pages. The import matches hOCR pages to pages by position, checks each page box against the target page’s pixel size, and replaces only the text layer as one undo step. Pages without OCR are exported as empty page boxes so they keep their position; importing an empty page box clears that page’s text layer. Annotations are not part of hOCR and are not transferred. The TIFF or other image files must be imported in the old session’s page order, because a wrong-order page of the same size cannot be detected from its page box.

A session with no recorded actions opens as an empty document, and a session whose latest action removed every page reopens empty. The first page scanned into a restored session is recorded as its first undo step.

Dependencies

Package names below are the Debian package names. Equivalent packages for other distributions are given in the wheel file installation instructions.

Required

• gir1.2-gdkpixbuf-2.0
• gir1.2-gtk-4.0
• imagemagick
• img2pdf
• libtiff-tools
• ocrmypdf
• poppler-utils
• python3-gi
• python3-gi-cairo
• python3-iso639
• python3-pil
• python3-sane
• python3-tesserocr

Optional

• djvulibre-bin
• jbig2 (for JBIG2 PDF compression)
• qpdf (for PDF encryption)
• unpaper
• xdg-utils

Development

• python3-pytest-mock
• python3-pytest-cov
• python3-pytest-timeout
• python3-pytest-xvfb

Download, Installation & Removal

Requirements

• Python 3.10 or later.
• GTK 4.6 or later and PyGObject 3.42 or later.
• The dependencies listed below. They are installed automatically when installing from a wheel or with uv, but must be installed manually when running from a tarball or the repository.

Ubuntu 22.04 (jammy), with GTK 4.6 and PyGObject 3.42, is supported until 2027-06. Ubuntu 24.04 (noble) and newer toolchains are also supported.

Packaged installs

Debian-based

• Debian sid should automatically have the latest version.

  sudo apt update
  sudo apt install scantpaper

• Ubuntu users can use the PPA:

  sudo apt-add-repository ppa:jeffreyratcliffe/ppa
  sudo apt update
  sudo apt install scantpaper

In either case to remove scantpaper afterwards:

sudo apt remove scantpaper

From Fedora RPM packages

Download the RPM files from Github and install them all in a single transaction:

sudo dnf install *.rpm

The release provides scantpaper together with locally rebuilt versions of python3-sane, python3-tesserocr, python3-cysignals and python3-iso639 (packaged as RPMs because the Fedora versions do not satisfy scantpaper’s dependencies, and with a python3-sane patch for Epson epsonscan2 scanners). scantpaper requires these packages by their python3dist(...) names, so they must be installed in the same transaction: installing scantpaper first fails with nothing provides python3.14dist(...).

To remove scantpaper afterwards:

sudo dnf remove scantpaper

Weak dependencies such as qpdf are installed by default, but can be skipped with --setopt=install_weak_deps=0, in which case the optional features they provide (e.g. PDF encryption) are disabled.

From openSUSE RPM packages

Download the Leap 16.0 RPM files - the ones with the .lp160.1 release suffix - from Github into a directory of their own and install them all in a single transaction:

sudo zypper install --allow-unsigned-rpm *.rpm

The release provides scantpaper together with locally rebuilt, Leap 16.0 versions of the Python dependencies that the distro does not package (python313-sane, python313-tesserocr, python313-cysignals, python313-python-iso639, python313-pdfminer.six, python313-img2pdf, python313-ocrmypdf, python313-setuptools-scm and python313-hatch-vcs, with a python313-sane patch for Epson epsonscan2 scanners).

Installing the set replaces any same-named packages you already have from repo-oss or from a third-party repository such as devel:languages:python. That is intended: the rebuilt packages carry fixes the repository builds do not, the python313-sane snap() patch being the important one. Three of them are versioned so that a later system update keeps your build instead of reverting it to the distribution’s.

To remove scantpaper afterwards:

sudo zypper remove scantpaper

The RPMs are unsigned (the workflow builds them from source), hence the --allow-unsigned-rpm. That is the only flag needed. PostScript export needs tiff2ps, which is not packaged on Leap 16.0, so the PostScript save type is not available there.

Tumbleweed

Download the Tumbleweed RPMs - the files without the .lp160.1 release suffix - into a directory of their own and install them in one transaction:

sudo zypper install --allow-unsigned-rpm *.rpm

Only the patched python-sane is rebuilt; every other Python dependency comes from the Tumbleweed repositories. The patched package replaces Tumbleweed’s unpatched 2.9.2 build, so Epson epsonscan2 scanners get the snap() fix. A later Tumbleweed python-sane release will sort above it until the fix is upstreamed. PostScript export has the same limitation as Leap: tiff2ps is not packaged on Tumbleweed (tiff 4.7.2), so the PostScript save type is not available.

From a wheel file

Download .whl from Github.

# Install the C-libraries that pip cannot handle:
# For Debian/Ubuntu
sudo apt update
sudo apt install libgirepository-2.0-dev libcairo2-dev libsane-dev pkg-config python3-dev gir1.2-glib-2.0
# For Fedora
sudo dnf install gobject-introspection-devel cairo-devel sane-backends-devel pkgconf-pkg-config python3-devel
# For openSUSE
sudo zypper install gobject-introspection-devel cairo-devel sane-backends-devel pkgconf-pkg-config python3-devel
# For Arch
sudo pacman -S gobject-introspection cairo sane pkgconf python
# For Homebrew
brew install pygobject3 gobject-introspection cairo sane-backends pkg-config
# Possibly upgrade pip
python3 -m pip install --upgrade pip
# Install from the wheel file, automatically including python dependencies
pip install scantpaper-x.x.x-py3-none-any.whl

If you haven’t already, you will then probably have to add ~/.local/bin to your path in order to find the new executable, after which you can start it with:

scantpaper

To then remove it:

pip uninstall scantpaper

With uv

To install the runtime dependencies with uv:

uv sync

or with the additional development dependencies:

uv sync --extra test

After which you can start it with:

uv scantpaper

From source

From a tarball

Download .tar.gz from Github. After installing the dependencies listed above:

tar xvfz scantpaper-x.x.x.tar.gz
cd scantpaper-x.x.x
PYTHONPATH=src python3 -m scantpaper.app

From the repository

Browse the code at Github. After installing the dependencies listed above:

git clone https://github.com/carygravel/scantpaper.git
cd scantpaper
PYTHONPATH=src python3 -m scantpaper.app

In either of the above two cases, just delete the source directory to remove it.

Support

• Mailing lists:
  • gscan2pdf-announce (announcements)
  • gscan2pdf-help (general support)

Reporting Bugs

• Please read the FAQs first.
• Report bugs preferably against the Debian package or Debian Bugs.
• Alternatively, use the Github issue tracker.
• Include the log file created by scantpaper --log=log with your report. On exit the log is compressed to log.xz, so submit that file.

Translations

scantpaper is partly translated into several languages. We’d like to organise the translations on Launchpad Rosetta — but if that’s too much effort for you, attaching a .po file to an issue or a merge request is just as welcome.

Missing strings are seeded as #, fuzzy (needs review) entries and fall back to English until a translator confirms them, so only confirmed translations reach users. The catalogs live in the po/scantpaper/ template directory, one po/scantpaper/<language>.po per language next to the generated po/scantpaper/scantpaper.pot — the layout Rosetta expects, so the directory can be tarballed and uploaded as-is.

• Scanner option translations come from sane-backends. Contribute via the sane-devel mailing list or SANE project.

To test updated .po files:

python3 dev/compile_mo.py --src po/scantpaper --out src/scantpaper/locale --domain scantpaper
PYTHONPATH=src python3 -m scantpaper.app --log=log --locale=locale

Set locale variables as needed (e.g., for Russian):

LC_ALL=ru_RU.utf8 LC_MESSAGES=ru_RU.utf8 LC_CTYPE=ru_RU.utf8 LANG=ru_RU.utf8 LANGUAGE=ru_RU.utf8 PYTHONPATH=src python3 -m scantpaper.app --log=log --locale=locale

Checking for upstream translations

python3 dev/check_launchpad.py

reports whether anything has moved on Launchpad since the last sync, by comparing the current state of the translation template and of every language against the committed baseline po/launchpad-state.json. Two triggers are watched: the template (the message set / .pot), read from the Launchpad JSON API with a conditional request, and each language’s last-changed time, which exists only on the public series translation page. A language is reported when its last-changed time is newer than the baseline, when a new locale appears, or when the template changes.

It also reports the reverse direction: whether the local template generated from source is stale relative to the last pot uploaded to Launchpad. The local message set is compared against the uploaded_pot fingerprint recorded in the baseline; a difference (new, removed or changed strings) is reported as Local pot is stale: naming the added and removed strings, and the exit code is unaffected unless --fail-on-change is given.

The check is advisory by default: it exits successfully whether or not it found changes, and it never downloads a catalog (Launchpad requires a login to export one). --fail-on-change makes a detected change non-zero for CI. The one always-non-zero outcome is a state the check could not read, so a broken check is loud rather than mistaken for “nothing new”. The committed baseline is advanced deliberately, after the catalogs have been synced:

python3 dev/check_launchpad.py --update

After a new template has been uploaded to Launchpad, record its message set so the next run no longer reports the pot as stale:

python3 dev/check_launchpad.py --record-pot

The translations-check workflow runs the same check daily and on demand (workflow_dispatch), and files one labelled issue when it detects upstream changes and a separate one when new strings are ready to upload.

Validating catalogs and source strings

python3 dev/check_po.py

This needs xgettext, msgcat and msgfmt on PATH. intltool is not needed: the translatable strings in the .ui files are extracted with the GtkBuilder ITS rules vendored in dev/, so the template does not depend on the gettext version installed. CI runs the same script via test_po_files.py. dev/check_po.py compares those vendored rules against the copies gettext installed and reports any drift as an advisory.

Errors fail the script; advisories are printed but do not. A release is therefore never blocked by an advisory, but they are worth reading.

The hard checks catch defects that break a catalog or a translation: a bad format specifier, an unbalanced or renamed placeholder, a plural entry with the wrong number of forms, a Plural-Forms header that disagrees with CLDR, an obsolete-entry count over the catalog’s ceiling, and counted strings written as %(plural)s rather than ngettext.

The advisories fall into two groups. Source-side ones report strings that are hard to translate, so they are best fixed by rewording: a concatenation fragment that fixes English word order into a translatable string, a msgid differing from another only by case, a doubled space, a msgid over 200 characters, and a μs written with GREEK SMALL LETTER MU. Catalog-side ones report mistranslations that only a native speaker can judge: a msgstr copied verbatim from a different msgid (the signature of a bad fuzzy match), and two sibling controls such as Select Odd and Select Even sharing one translation.

dev/check_po.py has a --pot option that points it at an existing template instead of generating one, which the test suite uses to exercise the source-side checks.

FAQs

Why isn’t option xyz available in the scan window?

It may not be supported by SANE or your scanner. If you see it in scanimage --help but not in scantpaper, send the output to the maintainer.

How do I scan a multipage document with a flatbed scanner?

Enable “Allow batch scanning from flatbed” in Preferences. Some scanners require additional settings. Switching to the flatbed starts # Pages at one page; choose a finite count, or choose All to scan until you cancel.

How do I scan a book?

Cut off the binding and scan every sheet one at a time, always aligning the untorn outer edge to the same rim. Scan in landscape to save time, set the rotation for the front pages in the Postprocessing tab, then tick Alternate rotation every 2nd page so the backs (which are rotated 180 degrees relative to the fronts) can be scanned too without manual rotation. For one-page-at-a-time flatbed scanning, the toggle remains active across consecutive scans so every second scan is flipped automatically; turn it off to reset parity.

Why is option xyz ghosted out?

The required package may not be installed (e.g., xdg-email, unpaper, imagemagick).

Why can I not scan from the flatbed of my HP scanner?

Set “# Pages” to “1” and “Batch scan” to “No”.

Why is the list of changes not displayed when updating in Ubuntu?

Only changelogs from official Ubuntu builds are shown.

Why can’t scantpaper find my scanner?

If the scanner is remote and not found automatically, specify the device:

scantpaper --device <device>

How can I search for text in the OCR layer?

Use pdftotext or djvutxt to extract text. Many viewers support searching the embedded text layer.

How can I change the colour of the selection box or OCR output?

The selection rectangle is styled with the .rubberband CSS class. scantpaper ships a default rule for it that follows the active GTK theme, so the selection is visible out of the box. To use a colour of your own, create or edit ~/.config/gtk-4.0/gtk.css (a user stylesheet overrides the application default):

.rubberband {
    border: 1px solid #2a76c6;
    background-color: rgba(42, 118, 198, 0.2);
}

#scantpaper-ocr-output {
    color: black;
}

What’s in a name?

“scant” (https://en.wiktionary.org/wiki/scant) in this sense means “short (of)”, as I am trying to digitalise my paperwork, and I liked the play on “scan”.

Known Limitations

PDFs larger than 2 GiB

Saving more than approximately 250 uncompressed scanned pages (at 300 dpi, 8-bit grayscale) produces a PDF exceeding 2 GiB. At that size, three tools in the save pipeline overflow 32-bit file offsets and produce truncated or corrupt output:

Scantpaper now estimates the output size before conversion and refuses to save when it would exceed 2 GiB, showing an error message suggesting fewer pages.

When updating dependencies, re-test by saving ~250 high-resolution uncompressed pages (e.g., 7000×5000 px grayscale TIFFs) and verifying the output PDF opens correctly in a PDF viewer. Note that Ghostscript’s 64-bit integer support (needed for >2 GiB files) is build-dependent — see the Ghostscript documentation on word size.

See Also

• gscan2pdf (the Perl predecessor of scantpaper)
• XSane
• Scan Tailor

History

I started writing gscan2pdf as a Perl & Gtk2 project in 2006. Version 2 switched to Gtk3, but kept the basic software architecture. This stored the pages as temporary files with hashed names, which had a couple of major disadvantages:

• Difficult to support PDF/A directly
• It was impossible to create documents with more than a few hundred pages, as it ran out of open file handles.
• In the event of a crash, it was tedious to recreate the document from the image files.
• AFAIK, Perl’s support for Gtk4 never extended beyond that provided by introspection.

Therefore I decided in 2022 to completely rewrite gscan2pdf in Python and renamed it for v3 scantpaper. The rewrite:

• Supports PDF/A by using ocrmypdf to write PDFs
• Stores all session data in a single Sqlite database
• Should be simple to migrate to Gtk4

See also the changelog for detailed release notes.

Author

Jeffrey Ratcliffe (jffry at posteo dot net)

Thanks To

• All contributors (patches, translations, bugs, feedback)
• The SANE project
• The authors of img2pdf and ocrmypdf, without which this would have been much harder.

Contributing

See contributing.

Donate

License

Copyright © 2006–2026 Jeffrey Ratcliffe jffry@posteo.net

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License v3 as published by the Free Software Foundation.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.