How is PDF compression being done?

I want you to tell me about how are the doing the pdf compression

Perfect! Now I have a comprehensive understanding of the PDF compression implementation. Let me create a detailed explanation for you.

##PDF Compression in Stirling-PDF

Based on my review of the source code, here's how Stirling-PDF implements PDF compression:

###Multi-Strategy Compression Approach

The compression system uses a layered approach with multiple compression techniques applied sequentially:

####1. Image Compression (Primary Method)

The system performs aggressive image optimization:

• Image Detection & Deduplication: Scans all pages to find images (both direct and nested within form XObjects)
• Unique Image Identification: Uses sophisticated hashing (MD5 on pixel data + metadata like dimensions, color space, bits per component, masks) to identify duplicate images
• Smart Replacement: When the same image appears multiple times, it's compressed once and all instances are replaced with the compressed version

Compression Process:

• Resizing: Images are scaled based on optimization level (28%-98% of original size)
• Grayscale Conversion: Optional conversion from color to grayscale
• JPEG Quality Control: Quality is adjusted per optimization level (92% for level 1 down to 35% for level 9)
• Optimization Parameters:
  • Huffman table optimization
  • Progressive JPEG encoding
  • Bicubic interpolation for high-quality resizing

Example from code:

// Scale factors by optimization level
case 1 -> 0.98;  // 98% of original
case 5 -> 0.68;  // 68% of original
case 9 -> 0.28;  // 28% of original (most aggressive)

// JPEG quality by level
case 1 -> 0.92f;  // 92% quality
case 9 -> 0.35f;  // 35% quality

####2. Ghostscript Compression (Level 6+)

For optimization level 6 and above:

• Uses Ghostscript's pdfwrite device for deep PDF optimization
• Applies different PDF settings based on level:
  • Levels 1-2: /prepress and /printer (highest quality)
  • Levels 3-5: /ebook and /screen (balanced)
  • Levels 6-9: /screen with aggressive image resolution downsampling (72-150 DPI)
• Additional features:
  • Detects and removes duplicate images
  • Compresses fonts
  • Subsets fonts (removes unused glyphs)
  • CMYK to RGB conversion at level 7+

####3. QPDF Recompression

Always applied (when enabled):

• Normalizes PDF structure and content streams
• Recompresses existing compression streams
• Generates object streams to group related objects
• Optional linearization for web viewing
• Advanced features at high levels:
  • JPEG quality optimization (78% → 34% quality)
  • Zopfli compression on Unix systems (better than standard zlib)
  • Decode/encode optimization

####4. Line Art Conversion

Optional conversion using ImageMagick:

• Converts images to high-contrast line art (useful for scanned documents)
• Configurable threshold (0-100) and edge detection strength (1-3 levels)
• Typically reduces file size significantly for document scans

###Optimization Levels (1-9)

Level 1: 98% resizing, 92% JPEG quality   → Minimal compression, best quality
Level 5: 68% resizing, 72% JPEG quality   → Balanced
Level 9: 28% resizing, 35% JPEG quality   → Maximum compression

###Auto-Mode (Size-Based Optimization)

The system can automatically determine optimization level based on a target size:

Target size:    Optimization Level Determined:
> 90% of original  → Level 1
> 80% of original  → Level 2
> 60% of original  → Level 4
> 30% of original  → Level 5
> 15% of original  → Level 7
< 10% of original  → Level 9

If target size isn't met, the system iteratively increases the optimization level and retries.

###Key Features

1. Duplicate Elimination: Same image compressed once, reused everywhere

2. Nested Image Support: Handles images inside form XObjects (complex PDFs)

3. Smart Minimum Thresholds:

   • Won't resize images already < 400x400 pixels (unless grayscale conversion)
   • Won't compress if reduction is < 5%

4. Fallback Logic: If optimized file is larger than original, keeps the original

5. Comprehensive Statistics: Logs compression ratios, bytes saved, number of images processed

###Processing Flow

1. Line Art Conversion (if requested)
   ↓
2. Ghostscript Compression (if level ≥ 6 and enabled)
   ↓
3. QPDF Recompression (if enabled)
   ↓
4. Image Compression (if level ≥ 4 or grayscale enabled)
   ↓
5. Check if target size met (auto-mode)
   → If not: increment level and retry

The system is quite sophisticated—it uses multiple compression techniques in layers, intelligently deduplicates images, and can automatically adjust compression intensity to hit a target file size!