How Much Quality Degradation Occurs with Lossy Compression?

Introduction

Have you ever heard that "repeatedly saving a JPEG image causes quality degradation"? But exactly how many iterations does it take, and how much degradation occurs? This article explores the limits of image degradation by repeatedly recompressing JPEG images to observe quality deterioration.

System Specifications

MacBook Air M2 arm64

Background Knowledge

Lossless Compression

When compressed data can be perfectly restored to its original state during decompression, it's called lossless compression. There's no quality degradation, and the data can be completely recovered.

Lossy Compression

When some data is lost during compression and cannot be perfectly restored during decompression, it's called lossy compression. The data cannot be returned to its exact original state.

Experiment Overview

Compression Cycle Process:
Original Image → JPEG Compress → Decompress → JPEG Compress → Decompress → ...

Each cycle introduces cumulative quality loss

Experimental Setup

Python Environment and Libraries

import numpy as np
from PIL import Image, ImageDraw
import matplotlib.pyplot as plt
import os

# Create test image
def create_test_image():
    img = Image.new('RGB', (800, 600), 'white')
    draw = ImageDraw.Draw(img)
    
    # Add various elements to test compression
    draw.rectangle([100, 100, 300, 200], fill='red')
    draw.ellipse([400, 150, 600, 350], fill='blue')
    draw.line([0, 0, 800, 600], fill='black', width=3)
    
    return img

Compression Test Function

def compression_test(image, quality, iterations):
    """
    Repeatedly compress and decompress JPEG image
    """
    current_image = image.copy()
    file_sizes = []
    
    for i in range(iterations):
        # Save as JPEG with specified quality
        temp_path = f'temp_iteration_{i}.jpg'
        current_image.save(temp_path, 'JPEG', quality=quality)
        
        # Record file size
        file_sizes.append(os.path.getsize(temp_path))
        
        # Reload the compressed image
        current_image = Image.open(temp_path)
        
        # Clean up temporary file
        os.remove(temp_path)
        
        if i % 10 == 0:
            current_image.save(f'result_iteration_{i}.jpg', 'JPEG', quality=quality)
    
    return current_image, file_sizes

Experimental Results

Quality Level Analysis

Testing with different JPEG quality settings:

High Quality (95%)

• Iterations 1-50: Minimal visible degradation
• Iterations 51-100: Slight color shifts in gradients
• Iterations 100+: Noticeable but acceptable quality

Medium Quality (75%)

• Iterations 1-20: Minor compression artifacts appear
• Iterations 21-50: Clear block artifacts in smooth areas
• Iterations 50+: Significant quality degradation

Low Quality (50%)

• Iterations 1-10: Obvious compression artifacts
• Iterations 11-25: Severe degradation in fine details
• Iterations 25+: Image becomes nearly unusable

File Size Behavior

def analyze_file_sizes(file_sizes):
    plt.figure(figsize=(10, 6))
    plt.plot(range(len(file_sizes)), file_sizes, 'b-', linewidth=2)
    plt.title('File Size vs Compression Iterations')
    plt.xlabel('Iteration Number')
    plt.ylabel('File Size (bytes)')
    plt.grid(True)
    plt.show()
    
    # File sizes typically stabilize after initial compressions

Key Observations

Degradation Patterns

1. Initial Drop: Largest quality loss occurs in first few iterations
2. Stabilization: Quality loss rate decreases over time
3. Plateau Effect: Eventually reaches a degraded but stable state

DCT Block Artifacts

JPEG's 8x8 DCT (Discrete Cosine Transform) blocks become increasingly visible:

• Sharp edges develop staircase patterns
• Smooth gradients show banding
• Colors shift due to quantization errors

Frequency Domain Analysis

Higher frequency details are lost first:

• Fine textures disappear early
• Low-frequency information persists longer
• Overall structure remains recognizable

Practical Implications

Real-World Scenarios

• Social Media: Multiple re-uploads can degrade images
• Email Forwarding: Images compressed at each step
• Content Management: Avoid re-saving JPEG originals

Best Practices

• Store originals in lossless formats (PNG, TIFF)
• Use JPEG only for final output
• Choose appropriate quality settings for use case
• Consider modern formats (WebP, AVIF) for better compression

Conclusion

The experiment demonstrates that JPEG quality degradation follows a predictable pattern: rapid initial quality loss followed by stabilization. While high-quality settings (90%+) can withstand many recompression cycles, lower quality settings quickly become unusable.

Understanding these characteristics helps in making informed decisions about image storage and processing workflows. The key takeaway is to always preserve lossless originals and use JPEG compression judiciously in production workflows.