OCR.space OCR for Bob

English | 简体中文

OCR.space OCR for Bob is an OCR plugin that connects Bob to the OCR.space API. It supports OCR.space Engine 1, 2, and 3, Bob image translation bounding boxes, and coordinate-based right-to-left vertical reading-order reconstruction.

• OCR.space Engine 1, 2, and 3.
• Bob OCR and Bob original image translation support through supportBoundingBox().
• Word-level bounding boxes converted from OCR.space pixels to Bob normalized [0, 1] coordinates.
• Vertical right-to-left page mode: columns are read top-to-bottom, then right-to-left.
• Configurable OCR language policy, layout policy, overlay, scale, orientation detection, and table mode.
• Build script that creates a .bobplugin package and updates appcast.json.

1. Download OCRSpaceOCR-v0.1.0.bobplugin from the GitHub Releases page.

2. Double-click the file to install it in Bob.

3. Open Bob preferences, add and enable the OCR.space OCR service under the corresponding features, and configure your OCR.space API Key:

   • For Screenshot Translation: Add the plugin's service to the list at Translation > Translation Settings > Services > Text Recognition.
   • For Independent OCR: Add the plugin's service to the list at OCR > OCR Settings.

   Tip: You can add the service multiple times with different settings (e.g., different engines or layout modes) to quickly switch between them depending on your needs.

Create or copy an API key from OCR.space Free OCR API. (For PRO plans, visit the main API page). The plugin sends the key in the apikey HTTP header, not in the URL. You can enter multiple keys separated by commas; the plugin chooses one randomly for each request.

OCR.space Engine 3 has the best OCR accuracy but is slower. Engine 1 is the default because it is the fastest and its overlay coordinates are the most precise, which makes it ideal for Bob's image translation bounding boxes and text selection.

For Japanese vertical pages, Engine 3 sometimes returns high-quality ParsedText but only coarse overlay blocks (which the plugin now handles by approximating the bounding boxes). If you need strict character-level vertical coordinates for precise right-to-left reconstruction and selection, use Engine 1.

See docs/configuration.md for all options.

Important defaults:

• engine: 1
• layoutMode: auto
• overlay: true
• scale: false
• detectOrientation: false

The plugin automatically maps Bob's requested source language to OCR.space three-letter language codes. If Bob is set to auto-detect the language, Engine 2/3 will use OCR.space's auto-detection feature, while Engine 1 will fall back to Bob's detected language.

The plugin sends the image as OCR.space base64Image in a URL-encoded request body. This keeps language, OCREngine, and overlay fields in the same form payload and avoids Bob multipart requests losing non-file fields on some setups.

For vertical text, do not rely on ParsedText alone. This plugin reconstructs reading order from OCR.space TextOverlay.Lines[].Words[] coordinates when the overlay is character-level or short-word-level:

1. Flatten all words with boxes.
2. Cluster words by horizontal center into vertical columns.
3. Sort columns from right to left.
4. Sort each column from top to bottom.
5. Return each column as a Bob OCR text object with a normalized bounding box.

Known first-version limitations: ruby text can form extra narrow columns, page headers and page numbers can enter the reading order, and two-page scans are not split automatically. See docs/vertical-japanese.md.

./build.sh

The package is written to:

release/OCRSpaceOCR-v0.1.0.bobplugin

The script updates appcast.json with the package SHA256 and timestamp.

See docs/release.md. The package must not include .git, local_assets, release, old .bobplugin files, test fixtures, caches, or temporary files.

See docs/troubleshooting.md.

Common causes:

• Missing or invalid API key.
• OCR.space quota or size limits.
• Engine 1 selected with languageMode=auto and no Bob-detected supported language.
• Overlay disabled, which prevents Bob image translation boxes and vertical reconstruction.

MIT