OCR Extract Table

OCR Extract Table

Type: ocr-extract-table  ·  Package: OCR Activities v1.2.5  ·  Output: DataTable

Since: v1.2.0

Detects a table inside an image, grouping the recognized words into rows and columns by position, and returns it as a DataTable. Useful to extract a tabular listing from a scanned document (an invoice detail, a data table).

How it works

Runs OCR on the image (with preprocess = "auto" by default) and rebuilds the table from the coordinates of each word:

1. Rows: each recognized line is a row.
2. Columns: it looks for the horizontal positions where words repeatedly start across the document (grouping starts less than 25 px apart) and uses them as columns. A position counts as a column if at least 2 words start there (or as many as a quarter of the number of lines, in long documents).
3. Each word goes to the column that starts just before it; several words in the same cell are joined with a space.
4. It discards rows with fewer filled cells than minColumns (titles, footers, loose text).
5. With hasHeader = true, the first remaining row gives the column names.

Required parameters

Parameter	Type	Default	Description
path	string	—	Path to the image file (.png, .jpg, .bmp, .tiff).

Optional parameters

Parameter	Type	Default	Description
language	string	"eng"	Text language: Tesseract code (eng, spa, eng+spa) or regional tag (es-CO). See language.
preprocess	string	"auto"	Prior cleanup: none · auto · deskew · binarize. See preprocessing.
hasHeader	bool	true	Treat the first detected row as column names.
minColumns	int	2	Discard rows with fewer filled cells than this number. Values below 1 are treated as 1.

Output

Returns a DataTable with every cell as text:

• Columns: with hasHeader = true, the texts of the first row; an empty header is named ColN and a repeated one gets a suffix (Amount, Amount_2). With hasHeader = false, Col1, Col2…
• Rows: the other detected rows; cells with no word stay empty.
• No text or no valid rows: an empty DataTable (no columns or rows). Check it with rowCount(table) == 0.

Example

OCR Extract Table   path = scanPath   language = "eng"   minColumns = 3   → output: table
If   condition = rowCount(table) == 0
 └─ then:
      Throw   message = "No table was detected in the image"
For Each   items = table   itemVariable = row
 └─ activities:
      Log   message = str(row["Description"]) + " — " + str(row["Amount"])

Approximate detection

Rebuilding a table from an image is inherently approximate: it depends on the columns being well aligned and the scan being sharp. Review the result: adjust preprocess, minColumns and hasHeader, and validate the data. If the original document is digital (not scanned), prefer PDF or Excel.

Common errors

Message	Cause and fix
File not found.	The path does not exist on the robot’s machine.
'path' is required.	path is missing.
Tesseract language '…' not found at '…'. Available: [eng, spa].	The language code is not installed.
Could not read the image.	The file is not a valid image or is damaged (preprocessing is on by default).
Error reading a missing column in a later step	Column names come from OCR and may vary (accents, case). Check table.Columns or use hasHeader = false and access Col1, Col2.

Related activities

• OCR Get Lines — the raw lines and coordinates, for fine control.
• DataTable — filter, sort and summarize the extracted table.
• PDF Get Text — for documents with real text.