Online Learning-based Video Reconstruction for Adaptive Bitrate Streaming

Temporal reconstruction of degraded frames in poor channel conditions

Developed online learning-based methods to reconstruct video frames during poor channel conditions in adaptive bitrate streaming. The approach leverages the inherent temporal self-similarity in video sequences to recover frames that suffer from heavy compression and downsampling.

Problem Statement: In adaptive bitrate video streaming, channel degradation causes:

  • Increased compression artifacts
  • Spatial downsampling
  • Frame quality reduction

Solution Approach:

  • Temporal self-similarity analysis across video frames
  • Online learning framework for frame reconstruction
  • Exploitation of correlation patterns between good and degraded frames

Applications:

  • Streaming services (Netflix, Prime Video, YouTube)
  • Mobile video delivery
  • Real-time communication platforms

Technical Paper