Evidence-backed Windows OCR guide
How to extract a table from an image to Excel
Recognize a photographed or scanned table, correct rows and cells, and export a structured XLSX without losing the title or isolated words.
Recognize the image, open the layout editor, confirm the table boundary and reading order, verify each row and column, then export XLSX. Compare the worksheet dimensions and values with the source before using formulas or deleting the scan.
Check release statusKeyword focus: extract table from image to ExcelStart with geometry the OCR can see
Crop excess margins only when preview confirms the table remains intact. Deskew or correct perspective for photographed pages, but protect borderless content from aggressive cropping.
Separate the title from the table
A report title should remain a Text region while the grid becomes a Table region. Otherwise the title may become an extra row or the first data row may be lost.
Verify dimensions before Excel export
Count source rows and columns, compare header cells, then inspect numbers whose separators or decimal points are easy to misread.
- Add the image
- Correct perspective if necessary
- Recognize
- Edit layout
- Confirm row and column count
- Export XLSX
- Compare values and formulas
Keep a round-trip record
Save the project, reopen it, and confirm user-edited table structure remains stable. This protects a repeatable workflow from changes that only existed in one UI session.
First-party workflow evidence
Controlled input → settings → expected → actual
- Controlled input
- Quarterly Revenue Table with a title, three headers, and three data rows.
- Mode and settings
- Real OCR → automatic Text/Table regions → project save/reload → structured export.
- Expected result
- Table is 4×3; title stays outside; all 15 OCR words remain; manual Table survives round-trip.
- Actual 0.7.0 result
- Frozen 0.7 met every count, retained user-edited Table state, preserved 16 words after an isolated-word test, and kept PDF/A tags at 4 TR / 3 TH / 9 TD.
Inspect all 140 named test results · Verify this guide workflow · Verify artifact hashes
Failure modes and limits
Merged cells, faint borders, rotated headers, handwriting, and dense financial formats may need manual structure correction. OCR does not validate business meaning.
Always inspect names, codes, dates, totals, and negative signs. A confidence score helps prioritize review; it does not make an unchecked result authoritative. Stop when the processed preview removes content or when expected and actual structure differ.
Recovery and verification
Keep the image and .uvocr project. If worksheet structure is wrong, correct the Table region and regenerate; do not silently patch a spreadsheet whose source mapping is lost.
Record the input filename, page range, recognition language, processing settings, application version, output format, and a few known search terms or cell values. That record makes the result reproducible and explains what was checked.
Bound to one frozen build
These statements use the 346,624-byte executable with SHA-256 31E46DEC406F6FFF4A8CF68EB66E347234C2DB87EFAA32FAAC7A3EBF504C9784. The product audit passed; public signing did not.
Frequently asked questions
Can OCR keep rows and columns?+
Yes when the table is detected or corrected as structured Table content. Verify the dimensions before use.
Will it overwrite my Excel file?+
Outputs are written atomically and source files are not overwritten.
Can it export CSV instead?+
Yes. CSV and TSV are available when a flat table is sufficient; XLSX is better for workbook-oriented review.
Official and primary sources
Links were checked August 15, 2026. External-product statements are limited to the cited vendor or standards source.
UtiliVera OCR
Recognize locally. Verify visibly. Export deliberately.
The public download opens after trusted code signing; until then, these pages preserve evidence without starting the 30-day clock.
See the product →