DoublePageScanSplitter

DoublePageScanSplitter splits scans of double pages (book openings) into a left page (verso) and a right page (recto). It's a small tool for digitisation projects. Fully automatic splitting often fails on historical material, while splitting by hand in an image editor is slow. With DoublePageScanSplitter you mark the gutter of each scan with a couple of clicks and the tool does the rest.

The tool comes in two versions:

Version	File	Requirements
Browser app (recommended)	code/DoublePageScanSplitter.html	A modern web browser. No installation and no internet connection needed.
Jupyter notebook	code/DoublePageScanSplitter_v2.ipynb	Python 3, OpenCV, NumPy, pandas, Jupyter

Both versions use the same splitting logic and write the same output files.

How it works

For each double page, click on the gutter twice: once near the top and once near the bottom. The two points define the split line, so the tool also handles gutters that are slanted or skewed. The image is split along that line and the next image opens.

There are two split modes:

• rotate (default): the image is rotated around the middle of the gutter until the gutter is vertical, then it is split straight down. Both pages come out deskewed. Corners that have no image content after the rotation are filled with the fill colour (white by default).
• cut: the image is cut along the slanted line with no rotation and no resampling, so the original pixels stay untouched. The small triangle that belongs to the other page is filled with the fill colour.

If the gutter is almost exactly vertical (less than 0.1°), both modes make a plain vertical cut.

Browser app

Use it online: https://dhofu.github.io/DoublePageScanSplitter/

Even the online version processes everything in your browser and never uploads your images.

1. Open the online version, or download code/DoublePageScanSplitter.html and open it in your browser. Double-clicking the file works, and it runs offline.
2. Click Open images or Open folder, or drag JPG, PNG or TIFF files onto the window.
3. Optional, Chrome/Edge only: click Output folder… to write the split pages straight into a folder. In other browsers, or if you don't pick a folder, the results are offered as a ZIP download.
4. Choose the Mode (rotate or cut), the JPEG quality (used for JPEG output only), the Fill colour and, if you like, your name as Executor (it's recorded in timing.csv).
5. Click the gutter of each image twice. A magnifier next to the cursor helps you place the points precisely.
6. When you're done, click Finish & export.

Everything runs locally in your browser and no images are uploaded anywhere.

Controls

Input	Action
Left click ×2	Mark the top and bottom of the gutter (order doesn't matter)
Right click / S	Don't split this image
R / Esc	Reset the clicks on the current image
← / →	Previous / next image
Click in the file list	Go to that image (to redo a split, just click twice again)

Browser support

Browser	Input	Output
Chrome, Edge (and other Chromium-based browsers)	Files, folders, drag & drop	Directly into a folder, or ZIP
Firefox, Safari	Files, folders, drag & drop	ZIP download

Jupyter notebook

Installation

pip install -r requirements.txt

Usage

1. Open code/DoublePageScanSplitter_v2.ipynb in Jupyter.
2. In the configuration cell, set sourceDir (folder with your scans) and targetDir (output folder). You can also set MODE, FILL, MIN_ANGLE and JPEG_QUALITY there.
3. Run all cells. An OpenCV window opens for each image:
   • Left click twice: top and bottom of the gutter. The window moves on automatically after the second click.
   • Right click: don't split this image.
   • r: reset the clicks on the current image.
   • Esc or closing the window: stop collecting points. The images you've already marked are still processed.

Subfolders of sourceDir are processed recursively and their structure is mirrored in targetDir.

code/DoublePageScanSplitter_v1.ipynb is the original version, which takes one click per image and splits strictly vertically. It's kept for reference.

Image formats

Both versions read JPEG (.jpg, .jpeg), PNG (.png) and TIFF (.tif, .tiff). The split pages are saved in the same format and with the same extension as the source image:

Source	Output
JPEG	JPEG with the chosen quality (default 95)
PNG	PNG (lossless)
TIFF	TIFF, Deflate compressed (lossless)

Notes:

• Output is always 8-bit colour. 16-bit images are reduced to 8 bit and greyscale images are saved as RGB. Resolution (DPI) metadata is not carried over.
• Only the first page of a multi-page TIFF is read.
• Browsers can't read TIFF on their own, so the browser app has its own TIFF reader. It handles uncompressed, LZW, Deflate, PackBits and JPEG-compressed TIFFs (1/2/4/8/16 bit; grey, RGB, palette or CMYK; strips or tiles). It doesn't support CCITT fax compression (common in black-and-white TIFFs) or BigTIFF. Use the notebook for those files.

Output

For every split image NAME.jpg (or .png, .tif, …) the tool writes two files:

• NAME_L_verso.jpg: the left page
• NAME_R_recto.jpg: the right page

It also writes two CSV files that document the run:

splitValues_v2.csv: one row per image, with a header row. New runs are appended.

Column	Meaning
Filepaths	Path of the source image
x1, y1	Upper point of the gutter (pixels)
x2, y2	Lower point of the gutter (pixels)
angle	Gutter angle against the vertical, in degrees (positive = the bottom leans to the right)
mode	rotate or cut

For images that weren't split, the coordinate, angle and mode columns are empty.

timing.csv: one row per run, with no header. Each row is appended.

sourceDir, sourceDirFileCount, targetDir, targetDirFileCount, elapsedTime (s), elapsedTime (HH:MM:SS), date/time, executor

Sample data

data/images/ contains four sample double-page scans. data/split/ contains the results of splitting them with version 1 of the notebook, including the v1 log files (splitValues.csv holds one x value per image).

Repository structure

DoublePageScanSplitter/
├── code/
│   ├── DoublePageScanSplitter.html       # browser app
│   ├── DoublePageScanSplitter_v2.ipynb   # notebook, two-point split (current)
│   └── DoublePageScanSplitter_v1.ipynb   # notebook, one-click vertical split (legacy)
├── data/
│   ├── images/                       # sample double-page scans
│   └── split/                        # sample output
├── index.html                        # GitHub Pages entry point (forwards to the browser app)
├── requirements.txt
├── LICENSE
└── README.md

License

The code is released under the MIT License.