Visual saliency guided perceptual adaptive quantization based on HEVC intra-coding for planetary images.

Area of Science:

• Space exploration data compression
• Digital image processing
• Machine learning for visual quality

Background:

• Spacecraft missions require efficient data handling due to limited storage and bandwidth.
• Planetary image acquisition generates large datasets, necessitating onboard compression.
• Maintaining visual fidelity during compression is crucial for scientific analysis.

Purpose of the Study:

• To develop an efficient onboard image compression technique for planetary exploration.
• To reduce data volume while preserving subjective visual quality.
• To adapt High Efficiency Video Coding (HEVC) using perceptual information.

Main Methods:

• Proposed a perceptual adaptive quantization technique using Convolutional Neural Networks (CNN) and HEVC.
• Employed a modified residual network to automatically extract image saliency maps.
• Implemented a Coding Tree Unit (CTU) level Quantization Parameter (QP) adjustment based on saliency maps.
• Utilized quantitative metrics correlating with human perception for quality evaluation.

Main Results:

• The proposed algorithm adaptively determined QP values for HEVC intra-coding.
• Saliency maps guided a flexible and adaptive bit allocation strategy.
• Achieved improved visual quality compared to standard HEVC coding.
• Demonstrated a maximum bitrate reduction of 7.17%.

Conclusions:

• The CNN-based perceptual adaptive quantization effectively reduces bitrate for planetary images.
• The method successfully balances data compression with subjective visual quality preservation.
• This approach offers a viable solution for efficient data management in space missions.