Enhanced Standard Compatible Image Compression Framework Based on Auxiliary Codec Networks.

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Deep neural networks offer potential for image compression enhancement.
• Current methods include learnable codecs, postprocessing, and compact representation networks.
• Existing approaches often lack compatibility with standard codecs or optimal coding efficiency.

Purpose of the Study:

• To propose a novel image compression framework compatible with existing standards.
• To improve the learning efficiency of compact representation and postprocessing networks.
• To achieve superior coding efficiency in standard-compatible image compression.

Main Methods:

• Development of a standard-compatible image compression framework utilizing auxiliary codec networks (ACNs).
• ACNs are designed to mimic image degradation processes of existing codecs.
• Integration of ACNs to provide accurate gradients for compact representation networks.

Main Results:

• The proposed framework demonstrates effective and optimal learning for compact representation and postprocessing networks.
• The framework achieves substantial performance improvements over existing algorithms.
• Validation of the framework's effectiveness using JPEG and High Efficiency Video Coding (HEVC) standards.

Conclusions:

• The novel ACN-based framework offers a standard-compatible solution for enhanced image compression.
• This approach overcomes limitations of previous methods by enabling optimal network learning.
• The proposed method significantly advances the state-of-the-art in efficient, standard-compliant image compression.