Image Compression Using Stochastic-AFD Based Multisignal Sparse Representation.

Area of Science:

• Signal Processing
• Image Compression
• Machine Learning

Background:

• Adaptive Fourier Decomposition (AFD) is a signal processing technique utilizing Szegö kernel dictionaries.
• Stochastic Adaptive Fourier Decomposition (SAFD) extends AFD for multi-signal processing.
• Existing image compression methods face limitations in balanced performance and adaptability.

Purpose of the Study:

• To introduce the first SAFD-based general multi-signal sparse representation learning algorithm.
• To propose a novel image compression framework utilizing SAFD.
• To evaluate the performance of the proposed methods against state-of-the-art techniques.

Main Methods:

• Development and implementation of a SAFD-based sparse representation learning algorithm for multi-signal processing.
• Design and implementation of an image compression framework leveraging SAFD.
• Comparative analysis with 13 established compression methods (JPEG, JPEG2000, BPG, deep learning-based).

Main Results:

• The proposed SAFD-based methods demonstrate the best balanced performance among compared techniques.
• The compression framework achieves high efficiency and quality, adjustable by the decomposition level.
• The methods are based on single image adaptive sparse representation learning and require no pre-training.

Conclusions:

• The SAFD-based approach offers a robust and adaptable solution for multi-signal processing and image compression.
• The developed algorithm and framework show significant potential for advancing image compression technology.
• The method's solid mathematical foundation supports its viability as a core technology in the field.