Extract a Table from an Image in Zoho Sheet and Reconcile OCR

Quick answer

Use Data → Data from Picture, review and correct the extraction preview, and insert into a clear destination. Then compare every cell against the image. A plausible total is not proof that OCR recognized the correct records, signs or decimal points.

This is extraction into editable cells, not placing an image in or over cells.

Availability, format limits and privacy

The official web OCR guide documents JPEG/JPG, PNG and BMP input, a 10 MB maximum, and English images. Zoho’s release history records web availability in March 2025; earlier mobile availability does not establish desktop-app parity.

Use an existing authorized Zoho Sheet web account with edit permission on a private practice workbook. Zoho advertises a free individual version, but no feature-level free-account entitlement was verified for this tool. If Data from Picture is missing, stop and check current web help and your account’s feature access. No sign-in, trial or upgrade is part of this exercise.

Selecting an image for a web extraction sends that image to the service. Use only the synthetic table below, or another image you are authorized to upload. Crop away unrelated windows, names and other private information before selecting the file. Do not use a customer invoice as a convenient first test. This guide did not upload an image or run OCR.

Prepare the original source table

Create a clear image containing only this English table. For example, display the table in an application you already use and capture just the table, or print and photograph it square-on. Keep the text sharp, the complete headers visible and the four columns separated. Save as an allowed format under 10 MB.

Code	Units	UnitPrice	Amount
A017	2	12.50	25.00
B008	5	1.20	6.00
C104	1	-3.75	-3.75

This is the complete original fixture, not an OCR result. C104 is a synthetic credit adjustment with a negative price and amount. Do not silently remove its minus signs because the other records are positive.

Keep this text table and the original image for comparison. Before extraction, calculate the reference values independently:

• Three records, identified by A017, B008 and C104.
• Units: 2 + 5 + 1 = 8.
• Line amounts: 2×12.50 = 25.00; 5×1.20 = 6.00; 1×−3.75 = −3.75.
• Total amount: 25.00 + 6.00 − 3.75 = 27.25.

The codes deliberately mix letters and digits. Matching the amount total cannot detect B008 becoming BOO8 or one code being assigned to the wrong row.

Open extraction and inspect the preview

Create a private workbook named OCR-Practice and an empty sheet named Extracted. Record the workbook URL. Reserve A1:D4 for the extraction and keep the surrounding cells empty.

1. Open Data → Data from Picture and select the prepared image. The documented alternatives include drag-and-drop or pasting an image.
2. Use the image adjustment step to crop or rotate if needed, then continue with Next.
3. Inspect the editable extraction preview. Correct misrecognized cells against the original image, not against what you think a typical invoice should contain.
4. Choose the empty destination starting at Extracted!A1 and use Insert only after the preview has four columns, one header row and three data rows.

These steps come from the official web guide. Its preview also offers append controls, but this first test avoids appending to existing records. A clear destination makes missing, shifted or duplicated rows easier to detect.

Review all 16 displayed cells, including headers. Specifically compare the digit zero in B008, the decimal points, and both minus signs in the C104 row. A blank or unreadable source cell should remain unresolved until you can read it; do not invent a likely value to make the total match.

Reconcile the inserted cells

After insertion, confirm that Code, Units, UnitPrice and Amount occupy A:D with data in rows 2:4. If the layout is different, stop and reconcile the layout before using these addresses. Do not paste the expected answers over the extraction merely to make a test pass.

In a separate check area, use:

Check	Formula or manual action	Expected result
Units are numeric	=COUNT(B2:B4)	3
Prices are numeric	=COUNT(C2:C4)	3
Amounts are numeric	=COUNT(D2:D4)	3
Units total	=SUM(B2:B4)	8
Amount total	=SUM(D2:D4)	27.25
Record identities	Read A2:A4 against the image	A017, B008, C104

In F2 enter =B2*C2-D2 and fill through F4. Each difference should be 0, subject to ordinary tiny floating-point rounding. Investigate a materially nonzero difference rather than rounding a large discrepancy away. These checks compare recognized quantities and prices with recognized amounts; they supplement, not replace, the source-image comparison.

If COUNT returns fewer than three, some entries may still be text. Inspect the exact text and locale before converting it. Number formatting alone is not evidence that a string became a number. The number-formatting guide explains the distinction. Keep codes as identifiers; they are not quantities to sum.

Understand what each check misses

Two hypothetical failures illustrate why multiple checks matter. They are not observed OCR outputs:

• If C104’s amount becomes +3.75, the amount total becomes 34.75, which is 7.50 too high. The line-product check also exposes the sign mismatch if its price is still negative.
• If both the price and amount in that row lose their minus signs, the line-product difference is still zero. The original-image comparison and independent 27.25 total are needed to detect the error.

Likewise, swapping two codes while leaving all numbers unchanged preserves every numeric check. Compare complete records, not only totals or columns independently.

If a price such as 12.50 becomes 1250, do not “repair” it by dividing the whole column by 100. Other prices may be correct. Correct only the cells whose source evidence justifies a correction and retain a short record of what changed.

Retry without duplicating the data

If extraction quality is poor, improve the source image and use a fresh empty sheet for the next attempt. Do not append the same three records underneath the first attempt: that would produce six rows and a misleading total of 54.50 if every record duplicated correctly.

Keep the original and the new attempt separate until record identities and all values match. If the text cannot be read reliably, stop automated extraction and transcribe the small table manually from a legible original, marking that output as manual transcription rather than successful OCR.

Save and acceptance checks

Wait for the completed workbook to save, close it, and reopen the same URL. Recheck the three codes, numeric counts, negative adjustment, 8-unit total, 27.25 amount total and three line differences. Retain the source image separately so another reviewer can inspect the evidence.

Verification: Official web documentation was read on 21 October 2026. The original fixture and error-case arithmetic were independently checked. No image upload, actual OCR preview, extraction accuracy, free-account entitlement or save/reopen behavior was tested. No fabricated screenshot or extracted image output is included.