ocr-tool（文字识别工具）

OCR Tool Skill

Use Tesseract OCR to extract text from images, particularly useful for financial charts, announcements, reports, and screenshots containing Chinese and English text.

When to Use

✅ USE this skill when:

• Processing screenshots of financial charts or announcements
• Extracting text from images containing Chinese/English text
• Analyzing "公告全知道" or similar financial announcement images
• Processing images with tabular data or structured information
• Extracting text from charts, reports, or documents in image format

When NOT to Use

❌ DON'T use this skill when:

• Text files are already available (use read tool)
• PDF files (use other PDF extraction tools)
• Images without text content
• When Tesseract is not installed

Setup

# Verify Tesseract installation
tesseract --version

# Install language packs if needed (for Chinese)
# Windows: Download chi_sim.traineddata from https://github.com/tesseract-ocr/tessdata
# Place in: C:\Program Files\Tesseract-OCR\	essdata\

Basic Usage

Extract Text from Image

# Basic OCR (English)
tesseract image.png output.txt

# Chinese OCR
tesseract image.png stdout -l chi_sim

# Chinese + English OCR
tesseract image.png stdout -l chi_sim+eng

# Specify output format
tesseract image.png output -l chi_sim+eng pdf txt

Common Patterns for Financial Analysis

# Extract text from financial announcement images
tesseract announcement.png stdout -l chi_sim+eng | grep -E "公司|股份|增长|利润"

# Process multiple images
for img in *.png; do
    echo "=== $img ==="
    tesseract "$img" stdout -l chi_sim+eng
done

# Save OCR results
tesseract financial_chart.png financial_analysis.txt -l chi_sim+eng

Integration with OpenClaw

Example: Process Telegram Image Messages

# When receiving image messages via Telegram
# 1. Image is automatically downloaded to media directory
# 2. Use OCR to extract text
# 3. Analyze extracted content

# Find latest image
latest_img=$(ls -t "$HOME/.openclaw/media/inbound/"*.png | head -1)

# Extract text
tesseract "$latest_img" stdout -l chi_sim+eng

# Analyze for specific patterns (company names, financial data)
tesseract "$latest_img" stdout -l chi_sim+eng | grep -oE "#[^ ]+|【[^】]+】"

Example: Financial Announcement Analysis

#!/bin/bash
# analyze_financial_image.sh

IMAGE="$1"
OUTPUT="analysis_$(date +%Y%m%d_%H%M%S).txt"

echo "=== OCR Analysis Report ===" > "$OUTPUT"
echo "Image: $IMAGE" >> "$OUTPUT"
echo "Time: $(date)" >> "$OUTPUT"
echo "" >> "$OUTPUT"

# Extract text
echo "=== Extracted Text ===" >> "$OUTPUT"
tesseract "$IMAGE" stdout -l chi_sim+eng >> "$OUTPUT"

echo "" >> "$OUTPUT"
echo "=== Key Information ===" >> "$OUTPUT"

# Extract company names
echo "Company Names:" >> "$OUTPUT"
tesseract "$IMAGE" stdout -l chi_sim+eng | grep -oE "[A-Za-z0-9]+股份|[A-Za-z0-9]+科技|[A-Za-z0-9]+集团" | sort -u >> "$OUTPUT"

# Extract stock codes
echo "" >> "$OUTPUT"
echo "Stock Codes:" >> "$OUTPUT"
tesseract "$IMAGE" stdout -l chi_sim+eng | grep -oE "[0-9]{6}\.[A-Z]{2,4}" | sort -u >> "$OUTPUT"

# Extract financial metrics
echo "" >> "$OUTPUT"
echo "Financial Metrics:" >> "$OUTPUT"
tesseract "$IMAGE" stdout -l chi_sim+eng | grep -oE "同比增长[0-9.]+%|利润[0-9.]+亿元|增长[0-9.]+%" | sort -u >> "$OUTPUT"

echo "Analysis saved to: $OUTPUT"

Advanced Usage

Multiple Language Support

# Chinese Simplified
tesseract image.png stdout -l chi_sim

# Chinese Traditional
tesseract image.png stdout -l chi_tra

# Japanese
tesseract image.png stdout -l jpn

# Korean
tesseract image.png stdout -l kor

# Multiple languages
tesseract image.png stdout -l chi_sim+eng+jpn

Image Preprocessing (Improve Accuracy)

# Convert to grayscale (using ImageMagick)
convert image.png -grayscale Rec709Luma grayscale.png
tesseract grayscale.png stdout -l chi_sim+eng

# Increase contrast
convert image.png -contrast -contrast enhanced.png
tesseract enhanced.png stdout -l chi_sim+eng

# Remove noise
convert image.png -despeckle denoised.png
tesseract denoised.png stdout -l chi_sim+eng

Batch Processing

# Process all PNG images in directory
for img in *.png; do
    base=$(basename "$img" .png)
    tesseract "$img" "output_${base}.txt" -l chi_sim+eng
    echo "Processed: $img -> output_${base}.txt"
done

# Process with parallel (if available)
find . -name "*.png" -print0 | parallel -0 tesseract {} {.}.txt -l chi_sim+eng

Common Use Cases

1. Financial Announcements ("公告全知道")

# Extract key information from financial announcements
tesseract announcement.png stdout -l chi_sim+eng | \
    grep -A2 -B2 -E "公司|股份|增长|利润|合同|中标|收购"

# Find company mentions
tesseract announcement.png stdout -l chi_sim+eng | \
    grep -oE "#[^ ]+|【[^】]+】|([A-Za-z0-9\一-\龥]+股份)"

2. Stock Charts and Tables

# Extract stock data from charts
tesseract stock_chart.png stdout -l eng | \
    grep -E "[0-9]+\.[0-9]+|[0-9]+%"

# Process tabular data
tesseract table.png stdout -l chi_sim+eng | \
    awk 'BEGIN {FS="[[:space:]]{2,}"} {for(i=1;i<=NF;i++) printf "|%s", $i; print "|"}'

3. Document Screenshots

# Extract structured document content
tesseract document.png stdout -l chi_sim+eng | \
    sed -n '/^[0-9]\+\./p'  # Extract numbered items

# Extract headings
tesseract document.png stdout -l chi_sim+eng | \
    grep -E "^#|^【|^（"

Troubleshooting

Common Issues

1. Poor OCR accuracy

- Preprocess images (grayscale, contrast enhancement) - Use appropriate language packs - Ensure image resolution is sufficient (300 DPI recommended)

1. Missing Chinese characters

- Verify chi_sim.traineddata is installed - Use -l chi_sim+eng for mixed content - Check image quality and font clarity

1. Tesseract not found

- Install Tesseract via package manager - Add Tesseract to PATH environment variable - Verify installation with tesseract --version

Improving Accuracy

# Use custom configuration
tesseract image.png stdout -l chi_sim+eng --psm 6  # Assume uniform block of text
tesseract image.png stdout -l chi_sim+eng --psm 11  # Sparse text

# PSM modes:
# 3 = Fully automatic page segmentation, but no OSD (default)
# 6 = Assume a single uniform block of text
# 11 = Sparse text. Find as much text as possible in no particular order
# 12 = Sparse text with OSD

# Use OEM (OCR Engine Mode)
tesseract image.png stdout -l chi_sim+eng --oem 1  # LSTM only
tesseract image.png stdout -l chi_sim+eng --oem 2  # Legacy + LSTM
tesseract image.png stdout -l chi_sim+eng --oem 3  # Default

Performance Tips

• For batch processing, consider parallel execution
• Cache OCR results for repeated analysis
• Preprocess images to improve speed and accuracy
• Use appropriate PSM mode for image type

Integration Examples

With Python Scripts

import subprocess
import re

def ocr_image(image_path, lang='chi_sim+eng'):
    """Extract text from image using Tesseract"""
    result = subprocess.run(
        ['tesseract', image_path, 'stdout', '-l', lang],
        capture_output=True,
        text=True,
        encoding='utf-8'
    )
    return result.stdout

# Example usage
text = ocr_image('announcement.png')
companies = re.findall(r'#(\S+)', text)
print(f"Found companies: {companies}")

With Shell Scripts

#!/bin/bash
# analyze_financial_images.sh

analyze_image() {
    local img="$1"
    echo "Analyzing: $img"
    
    # Extract text
    text=$(tesseract "$img" stdout -l chi_sim+eng)
    
    # Extract key information
    echo "=== Summary ==="
    echo "Companies: $(echo "$text" | grep -oE '#[^ ]+' | tr '\
' ' ')"
    echo "Stock Codes: $(echo "$text" | grep -oE '[0-9]{6}\.[A-Z]{2,4}' | tr '\
' ' ')"
    echo "Financial Terms: $(echo "$text" | grep -oE '同比增长|利润|增长|合同' | sort -u | tr '\
' ' ')"
}

# Process all images
for img in "$@"; do
    analyze_image "$img"
    echo ""
done

Notes

• Tesseract works best with clean, high-contrast images
• Chinese OCR requires chi_sim/chi_tra language data files
• For financial charts with small text, ensure image resolution is sufficient
• Consider image preprocessing for better results with screenshots
• Always verify OCR results, especially for critical financial data