Image compression by self-organized Kohonen map

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Digital image compression is crucial for efficient storage and transmission.
• Existing standards like JPEG have limitations in compression efficiency and performance at high compression rates.
• Vector quantization (VQ) is a common technique in image compression.

Purpose of the Study:

• To present a new image compression scheme leveraging Kohonen's neural network.
• To explore the benefits of both vector quantization and topological properties of Kohonen's network for compression.
• To evaluate the performance of the proposed scheme against the JPEG standard.

Main Methods:

• Utilized Kohonen's self-organizing map (SOM) for vector quantization.
• Exploited the topological ordering property inherent in Kohonen's algorithm to enhance compression.
• Compared the proposed method with the Joint Photographic Experts Group (JPEG) standard.

Main Results:

• Achieved an approximate 80% increase in compression rate due to the topological property.
• Demonstrated superior performance (in terms of Peak Signal-to-Noise Ratio - PSNR) compared to JPEG at compression rates exceeding 30.
• The neural network-based approach offers significant advantages for high-rate image compression.

Conclusions:

• Kohonen's neural network offers a powerful approach for digital still image compression.
• The topological property of the network significantly boosts compression efficiency.
• The proposed scheme outperforms JPEG at higher compression ratios, indicating its potential for advanced image compression applications.