AI-Based Image Compression Using Genetic Algorithms

Oct 2025 – Dec 2025 · Brock University

Overview

This project formulates image compression as an optimization problem rather than relying on fixed interpolation methods. A block-based Genetic Algorithm (GA) is used to compress high-resolution RGB images (2000×1200) into compact 200×120 representations, which are later reconstructed through evolutionary optimization.

Challenges

  • Extremely large search space due to pixel-level representation
  • High computational cost of fitness evaluation
  • Preserving global structure under a 100:1 compression ratio
  • Block boundary artifacts during reconstruction

Actions Taken

  • Designed a block-based GA to reduce optimization complexity
  • Implemented pixel-level chromosome representation
  • Used MSE-based fitness with upscaling reconstruction
  • Applied selection, crossover, and mutation operators
  • Compared results against nearest, bilinear, and bicubic baselines

Results

The GA-based approach consistently preserved global image structure and dominant color regions across object-focused, portrait, and natural scene images. Fine-grained texture was partially sacrificed, reflecting the aggressive compression ratio and computational limits.

Object image compression result

Object-based image: original, GA-compressed (200×120), reconstructed.

Portrait compression result

Portrait image: global facial structure preserved, fine details degraded.

Natural scene compression result

Natural scene: large-scale structure retained, textures smoothed.

Evaluation Metrics

Performance was evaluated using Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index (SSIM), alongside visual inspection and baseline interpolation comparisons.

Resources