Low-Complexity Error Resilient HEVC Video Coding: A Deep Learning Approach.

Area of Science:

• Video Coding
• Deep Learning
• Network Robustness

Background:

• Error resilient video coding enhances streaming over error-prone networks.
• Existing methods suffer high computational complexity due to distortion prediction and brute-force search.
• This complexity limits practical application in real-time scenarios.

Purpose of the Study:

• To propose a Low Complexity Mode Switching based Error Resilient Encoding (LC-MSERE) method.
• To reduce the computational complexity of error resilient video encoding.
• To leverage deep learning for efficient coding unit partition prediction.

Main Methods:

• Developed multi-scale information fusion-based Convolutional Neural Networks (CNNs).
• Proposed spatial multi-scale information fusion CNN (SMIF-Intra) for intra coding unit prediction.
• Proposed spatial-temporal multi-scale information fusion CNN (STMIF-Inter) for inter coding unit prediction.
• Utilized image, coding, and transmission parameters for accurate partition prediction.

Main Results:

• Achieved rapid and accurate prediction of intra and inter coding unit partitions.
• Significantly reduced computation time for error resilient video encoding.
• Maintained acceptable video quality despite complexity reduction.

Conclusions:

• The LC-MSERE method effectively reduces encoder complexity.
• Deep learning approaches offer a viable solution for efficient error resilient video coding.
• The proposed CNN models accurately predict coding unit partitions, enhancing robustness.