OCR Activities

Version: 1.2.5  ·  ID: zoan-packages-ocr  ·  Author: Zoan Software

The OCR package (Optical Character Recognition) reads the text that appears inside an image: a scanned document, a photo of a document, a screenshot. It complements the PDF package: when a document is an image (it has no selectable text), the PDF activities cannot read it — but OCR can.

It uses the Tesseract 5 engine in LSTM mode, a widely used open-source text recognizer, with OpenCV-based image preprocessing. Spanish and English come bundled with the package: there is nothing else to install. Recognition runs on the robot’s machine (images are not sent anywhere) and the package works on Windows robots.

OCR is not perfect

OCR interprets an image, so its accuracy depends on the quality of that image: good resolution and sharp text give excellent results; a blurry, skewed or low-quality scan produces errors. Therefore:

• Use good-resolution images (300 DPI or more for scans).
• Turn on preprocessing (preprocess) to clean up difficult images.
• Validate the extracted data (for example, that a number has the expected format) — don’t assume the text is 100% correct.

Input formats

The activities that read files accept .png, .jpg, .bmp and .tiff images. For a scanned PDF, first convert its pages to images with PDF to Images and pass each image to OCR.

Text language

Almost every activity accepts a language parameter with the code of the language to recognize:

Value	Language
eng	English (default)
spa	Spanish
eng+spa	Several languages at once

It also accepts regional tags such as es-CO or en-US, which are mapped to the Tesseract code (es → spa, en → eng; also pt, fr, de, it and ca). An unrecognized tag is treated as English. A code that is not installed raises the error Tesseract language '…' not found, with the list of available ones.

Setting the right language greatly improves accuracy (accents, ñ, etc.). For Spanish text, always use language = "spa". To see which languages are available, use OCR Get Available Languages.

Image preprocessing

The preprocess parameter cleans up the image before recognizing it, which improves results on poor-quality scans. It is available in OCR Recognize File and the three structured-extraction activities:

Value	What it does
none	No preprocessing (default, except in OCR Extract Table, which uses auto)
auto	Full cleanup: grayscale, noise removal, binarization and deskew
deskew	Only straightens a skewed image (rotates it if the skew exceeds about 0.3°)
binarize	Converts to black and white with an adaptive threshold (improves contrast)

The original image is not changed: the cleanup happens in memory.

If recognition comes out wrong, try auto

preprocess = "auto" solves most problems with skewed or noisy scans. Start without preprocessing; if the text has errors, switch to auto.

Two sources: file or screen

• From an image file (.png, .jpg, .bmp, .tiff): OCR Recognize File and the structured-extraction activities.
• From the screen: OCR Recognize Screen Region (an area), OCR Screenshot (the whole screen) and OCR Find Text (in a region). These require an interactive desktop session (like the Desktop package): they capture what is visible at that moment.

Coordinates

The activities that return positions (OCR Find Text, OCR Get Lines) use pixels with the origin at the top-left corner of the analyzed image: the file’s, or on screen, the captured region’s. To get screen coordinates, add the region’s x/y.

Activities

Recognize text

Activity	Type	Output	What it does
OCR Recognize File	ocr-recognize	string	Reads the text of an image file
OCR Recognize Screen Region	ocr-recognize-region	string	Reads the text of a screen area
OCR Screenshot	ocr-screenshot	string	Reads all the text of a screen

Find and locate

Activity	Type	Output	What it does
OCR Find Text	ocr-find-text	object	Finds the position of a text
OCR Get Lines	ocr-get-lines	List<object>	Structured list of lines and words with coordinates

Structured extraction

Activity	Type	Output	What it does
OCR Extract Pattern	ocr-extract-pattern	Dictionary<string, string>	Extracts fields with regular expressions
OCR Extract Form Fields	ocr-extract-form	Dictionary<string, string>	Extracts text from named regions
OCR Extract Table	ocr-extract-table	DataTable	Extracts a table from an image into a DataTable

Utility

Activity	Type	Output	What it does
OCR Get Available Languages	ocr-get-languages	List<string>	Available languages

A typical flow

Read a scanned invoice and extract its number:

OCR Recognize File   path = imagePath   language = "spa"   preprocess = "auto"   → output: text
Set Variable   name = number   value = regexFind(text, "INV-\d+")

Next steps

• OCR Recognize File — read text from an image.
• OCR Extract Pattern — extract specific fields from a document.
• PDF — for PDFs with selectable text (no OCR).