DEV Community

nan nan
nan nan

Posted on Edited on

How to Extract Text from Multiple Screenshots at Once on Windows

#productivity #software #tools

(https://lijiannan1241.github.io/CaptureOCR/blog/extract-text-multiple-screenshots-windows)
If you’ve ever needed to pull six fields from an invoice, four error lines from a dialog, or twelve cells from a PDF, you know the pain: take a screenshot, run OCR, copy text, repeat. Ten times. Then paste into Excel and realize you lost track of which text came from where.

There’s a better way.

The Old Way (What You’re Probably Doing Now)
Press Win+Shift+S or open Snipping Tool
Drag around one field
Click “Text Actions” → “Copy all text”
Switch to Excel, paste
Repeat 9 more times
Hope you didn’t mix up row 3 and row 7
Windows Snipping Tool and PowerToys Text Extractor are great for one region. They return a single text blob. No numbering, no coordinates, no batch workflow.

The CaptureOCR Way
CaptureOCR is built for this exact scenario. Here’s the workflow:

  1. Start a Capture Session
    Press Ctrl+Shift+X (default hotkey) or click the tray icon. The screen dims and you enter selection mode.

  2. Drag Around Each Region
    Click and drag around the first field. A numbered badge appears: ①. Drag the second field: ②. Third: ③. Up to 10+ regions in one session.


![ ](https://dev-to-uploads.s3.us-east-2.amazonaws.com/uploads/articles/d6fcj5v4xokf0ml9ctfb.png)
Enter fullscreen mode Exit fullscreen mode

Regions are numbered automatically. Reorder or delete before recognition.

  1. Reorder or Delete (Optional)
    The region drawer shows all captures. Drag to reorder. Right-click to delete. The numbers update automatically — but internally each region keeps a stable ID so your export stays consistent.

  2. Run OCR
    Click “Recognize” or enable auto-OCR. Two modes:

Fast (tiny model, ~6.6 MB): ~78 ms per image, 49 languages, no Japanese
Accurate (small model, ~30 MB): ~258 ms per image, 50 languages including Japanese
Both run offline via ONNX Runtime. No cloud API calls.

  1. Export Structured Data This is where it pays off. Export options:

Format What You Get
Markdown table Region index, coordinates (x,y,w,h), line number, confidence, text
CSV (UTF-8 BOM) Same fields plus session_id, mode, elapsed_ms, model_hash — opens correctly in Excel
TXT Plain text blocks separated by region
.snipx session package ZIP with original images, Markdown, CSV, and full JSON with normalized boxes


very row carries its region index, coordinates, confidence, and text.
Real Example: Processing an Invoice
Say you have this invoice on screen:

Invoice number (top right)
Date (top left)
Vendor name (middle left)
Line items table (middle, 5 rows)
Total (bottom right)
Old way: 9 screenshots, 9 copy-pastes, manual cleanup in Excel.

CaptureOCR way:

Ctrl+Shift+X → drag 5 regions (invoice #, date, vendor, line items area, total)
Auto-numbered ①②③④⑤
Click “Recognize” → done in ~400ms total
Export CSV → open in Excel → every row labeled with region index and coordinates
No retyping. No “wait, which cell was this?” The coordinates tell you exactly where on screen each piece came from.

When This Matters Most
Support tickets: Error dialog + log line + customer message → one table
Research: Figure labels + table cells + citations from a PDF → traceable rows
Finance ops: Invoice fields → CSV → accounting system
QA testing: Multiple UI states captured once, exported for regression docs

Top comments (0)