Image Extract
feizhu66/image-extract
Extrai dados de imagens/PDFs para tabelas CSV/Excel sem OCR pago
Ver no GitHubSkill para Claude que extrai texto e dados de imagens e documentos digitalizados em tabelas estruturadas (CSV/Excel) usando visão nativa do Claude, sem custo adicional. Suporta processamento em lote, PDFs, múltiplas folhas Excel e documentos em chinês com marcadores de confiança.
Instalação
npx convez-ai feizhu66/image-extract1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Digitalizar e estruturar recibos, faturas e vouchers contábeis
- Extrair dados de formulários manuscritos e documentos digitalizados
- Processar em lote múltiplas imagens de documentos comerciais chineses
- Converter bundles de PDFs em planilhas Excel organizadas por categoria
Veredito

Solução prática e sem custos para extração estruturada de dados de imagens, especialmente útil para documentação contábil e formulários em chinês. Implementação clara com marcadores de confiança que evitam erros silenciosos.
README
A tradução para o português deste README ainda está na fila.
image-extract
Claude Code skill for batch extracting text and data from images and scanned documents into structured tables (CSV/Excel).
Features
- Zero cost — uses Claude's built-in vision, no paid OCR API needed
- Batch processing — handles multiple images and PDFs at once
- Auto PDF detection — PDFs are automatically converted to images for extraction
- Smart extraction — identifies document types: invoices, receipts, forms, contracts, handwritten notes
- Multi-sheet Excel — separate categories into different sheets (e.g., invoices vs. contracts)
- Confidence markers — unclear data marked with
[?]/[??]instead of guessed - Chinese support — works with Chinese documents, receipts, and forms
- Cross-document validation — flags mismatches between related documents in a bundle
Supported File Types
| Type | Extensions | Notes |
|---|---|---|
| Images | PNG, JPG, JPEG, BMP, TIFF, TIF | Processed directly |
| Auto-converted to images via PyMuPDF | ||
| Directory | folder path | Scans for all supported files |
Installation
# Clone the repo
git clone https://github.com/feizhu66/image-extract.git
# Copy to your Claude Code skills directory
# macOS/Linux:
cp -r image-extract ~/.claude/skills/
# Windows:
xcopy /E /I image-extract %USERPROFILE%\.claude\skills\
Dependencies
The skill uses Python libraries that may need to be installed:
# Required for PDF processing
pip install PyMuPDF
# Required for Excel output
pip install openpyxl
Usage
The skill triggers automatically when you mention image/data extraction. Just provide file paths:
Single image
帮我提取这张发票的数据: receipt.jpg
Multiple images
这几张收据整理成表格: img1.jpg, img2.jpg, img3.jpg
PDF document
处理这个PDF扫描件: contract.pdf
Directory of images
批量处理这个文件夹里的所有票据: C:\Documents\receipts\
Multi-document bundle with categories
这个8页PDF包含记账凭证、审批单、合同和发票,按类别分sheet导出Excel
Low-quality scans
这些扫描件有些字看不清,能提取多少算多少,看不清的标出来
Output Examples
Single receipt
Input: receipt.jpg (restaurant receipt)
Output: receipt_extracted.csv
| file | merchant | date | items | amount | tax | total |
|---|---|---|---|---|---|---|
| receipt.jpg | 海底捞 | 2024-01-15 | 火锅套餐 x1, 饮料 x2 | 240.00 | 15.60 | 258.00 |
Handwritten form with confidence markers
Input: form_scan.png (handwritten application form)
Output: form_extracted.xlsx
| file | field | value | confidence |
|---|---|---|---|
| form_scan.png | 姓名 | 张三 | high |
| form_scan.png | 电话 | 138xxxx1234 | high |
| form_scan.png | 地址 | 北京市朝阳区...[?] | medium |
Multi-sheet accounting bundle
Input: 8-page PDF with vouchers, approvals, contracts, invoices
Output: accounting_bundle.xlsx (4 sheets)
- Sheet 1 "记账凭证": voucher number, date, accounts, amounts
- Sheet 2 "付款审批单": approval number, applicant, payee, bank details
- Sheet 3 "合同信息": contract name, parties, service, amount
- Sheet 4 "发票信息": invoice code, date, buyer, seller, tax details
Manual Export Script
Use the bundled script to convert extracted JSON to CSV/Excel:
# Single sheet (flat data)
python scripts/export_table.py data.json output.xlsx
# Multi-sheet (categorized data)
python scripts/export_table.py data.json output.xlsx --multi-sheet
# Custom sheet name
python scripts/export_table.py data.json output.xlsx --sheet-name "发票数据"
# CSV output
python scripts/export_table.py data.json output.csv
Input JSON format
Flat data (single sheet):
[
{"merchant": "海底捞", "date": "2024-01-15", "amount": 258.00},
{"merchant": "星巴克", "date": "2024-01-16", "amount": 42.00}
]
Categorized data (multi-sheet):
{
"记账凭证": [
{"凭证编号": "记-2025-04-17", "金额": 9427.72}
],
"发票信息": [
{"发票代码": "25112000000060338877", "价税合计": 9522.00}
]
}
How It Works
User provides files
|
v
Detect input type
| |
v v
Image PDF
| |
| Convert to images
| (PyMuPDF, 200 DPI)
| |
v v
Read each image (Claude vision)
|
v
Extract text + data
|
v
Apply confidence markers
|
v
Validate (cross-doc checks for bundles)
|
v
Export to CSV or Excel
Error Handling
| Scenario | Behavior |
|---|---|
| Corrupted image | Skip, report which files failed |
| Blurry/unclear text | Mark with [?] or [??], extract what's readable |
| PDF conversion fails | Report error, continue with other files |
| Missing fields | Leave blank, don't guess |
| Numbers don't add up | Flag discrepancy for user to verify |
| openpyxl not installed | Fall back to CSV output |
| PyMuPDF not installed | Prompt user to install |
License
MIT