Strategies for enhancing deep video encoding efficiency using the Convolutional Neural Network in a hyperautomation...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Video Compression

Background:

• Three-dimensional (3D) video and depth video are increasingly important for immersive experiences.
• Efficient encoding of depth video is crucial for widespread adoption of 3D technology.
• Current video encoding methods face challenges in optimizing depth video compression.

Purpose of the Study:

• To propose an efficient deep video encoding method for 3D depth video.
• To integrate Convolutional Neural Networks (CNNs) with hyperautomation for enhanced encoding.
• To evaluate the performance of a novel variable-resolution intra-frame prediction technique.

Main Methods:

• Overview of CNN principles and hyperautomation concepts.
• Application of CNNs in the intra-frame prediction module of video encoding.
• Development of a CNN-based variable-resolution intra-frame prediction method using multi-level feature fusion.

Main Results:

• The proposed method achieved an average Bjøntegaard delta bit rate (BDBR) savings of 8.12% compared to the HEVC test platform (HTM-16.2).
• Viewpoint BDBR loss was minimal (0.15%), indicating stability and reliability in viewpoint coding.
• Superior performance in peak signal-to-noise ratio, structural similarity index, and perceptual quality metrics was observed.

Conclusions:

• The integrated CNN and hyperautomation approach offers a novel and efficient solution for 3D video compression.
• The variable-resolution coding technique significantly enhances depth video encoding efficiency.
• This research provides valuable insights for future advancements in AI-driven video encoding technologies.