How to Convert Image to Excel With OCR

A photo of a table is not a table. Until OCR makes it one. Here is how to turn any phone snap, screenshot or scan into a real Excel spreadsheet — in your browser, without uploading anything.

What "image to Excel" actually does

Two things have to happen: the words in the picture have to be read (OCR), and those words have to be placed into the right row and column of a spreadsheet. The first part — OCR — has improved dramatically in the last few years, and tools like Tesseract are now good enough to run inside a browser tab. The second part — table reconstruction — uses the bounding box of every recognised word to cluster them into rows by vertical position and columns by horizontal position. That is the same approach Camelot and img2table use under the hood, just running locally in JavaScript.

For a clean screenshot of a dashboard, this works almost flawlessly. For a phone photo of a printed receipt taken in good light, expect to fix one or two cells. For a glare-heavy, angled photo of a faded printout, expect to fix more — but you still save the time of retyping the whole table.

Step-by-step

Open Easy Press Pro's Image to Excel tool. Everything runs in your browser.

Step 1 — Take or choose the image

Supported formats: PNG, JPG, JPEG, WebP. iPhone HEIC photos need to be converted to JPG first using HEIC to JPG — both tools are free and browser-based.

Step 2 — Upload and let OCR run

Drop the image. The first time the tool runs OCR, your browser downloads a small (~10 MB) Tesseract language pack and caches it. Every subsequent run is fast. For a 1500×1000 photo, OCR typically finishes in 5–15 seconds.

Step 3 — Review the detected table

Every cell shows up in an editable preview grid. Click any cell to fix typos, merge two values into one, or correct an OCR mistake. Numbers, currency symbols and percentages are detected automatically and saved as numeric cells.

Step 4 — Download as XLSX or CSV

XLSX is the native Excel format with full number formatting. CSV is universal text — perfect for importing into databases, Google Sheets, or analytics tools. Either way, the file downloads instantly without any server roundtrip.

How to get the best OCR accuracy

• Lighting beats hardware. Even natural light. No glare patches. No shadows across the table. Window light works better than flash on a recent phone.
• Shoot straight on. Hold the camera parallel to the paper. Even small angles distort the columns at the edges, which hurts column detection.
• Crop tightly. Cut out everything that is not the table before uploading. Less noise → cleaner column boundaries.
• Resolution sweet spot is 1500–3000 px on the long edge. Below 800 px and OCR misses thin strokes. Above 4000 px and processing slows for no extra accuracy.
• Avoid heavy compression. A high-quality JPG (quality 85+) is fine. A WhatsApp-forwarded photo that has been compressed three times is harder for OCR.
• Use the editable preview. Glance at the result, fix 2–3 obvious cells, then export. That last step makes the difference between "almost useful" and "ready to send."

Real-world use cases

• Receipts and bills — photograph a paper receipt, get line items into Excel for expense reports or budgeting.
• Whiteboard photos from meetings — turn a brainstorm snapshot into a structured shareable spreadsheet.
• Screenshots from dashboards — when you can see the numbers but cannot export from the tool itself.
• Supplier price lists — vendors send PDF or photo catalogues; you need them in Excel to compare and re-price.
• Old paper records and printouts — digitise once, never retype.
• Forms and surveys photographed in the field — convert collected data into a working spreadsheet.

Limits to know about

OCR is not magic. Handwriting is unreliable. Tables stylised inside infographics with overlapping graphics are hard. Languages other than English need a different language pack. And for the same reason that compression is asymmetric — easy to do, hard to undo — a printed table photographed at a steep angle on a bad day will require more cleanup than a clean screenshot.

For those edge cases the editable preview is your friend. The point of this workflow is not "click and forget" — it is "click, glance, fix the bits OCR missed, done in 30 seconds." That is still vastly faster than typing a 20-row table by hand.

Source is a PDF, not an image?

Use PDF to Excel instead — same engine, tuned for PDF input. For text-based PDFs it skips OCR entirely and uses the document's own coordinates, which is faster and more accurate.