Abstract
Streaming video has largely increased, and the growing market of mobile entertainment and high-quality multimedia content over web sites can be expected to continue. Hybrid codecs like H.264 are widely used in modern multimedia applications. Motion compensation techniques are applied in most video coding schemes to compress the video data into acceptable sizes for limited transmission bandwidth. Video transmission over error-prone wired and wireless channels often suffers packet losses and bit errors which degrade video quality. The loss of a slice not only causes quality degradation itself but also allows errors to propagate into following frames. In this paper, a perceptual error protection model is proposed by utilizing a priority assignment schedule with Flexible Macroblock Ordering (FMO) and Automatic Repeat Request (ARQ). The prioritization is based on the properties of the human visual system and its influence on end-to-end quality. Macroblock (MB) coding mode, motion vectors, partition size, and bit consumption are utilized for the derivation of priority from spatial and temporal domains. Local motion properties and texture complexity are utilized for the derivation of perceptual saliency. A content-aware selective ARQ mechanism is then realized by integrating the proposed classification method. Results show that the proposed method outperforms former ARQ methods in both objective and subjective video quality.
Introduction
The surge in mobile video consumption has placed immense pressure on network infrastructure. Modern codecs like H.264 rely heavily on motion compensation, where frames are predicted based on previous data. While efficient for compression, this creates a vulnerability: a single lost packet can lead to error propagation, corrupting a sequence of frames.
Traditional error control mechanisms like Automatic Repeat Request (ARQ) often fail in real-time, low-latency scenarios because they attempt to retransmit all lost data indiscriminately. This can lead to bandwidth congestion and excessive delay. This paper introduces a "Texture and Motion-tracked Prioritization" (TMP) model designed to identify and protect only the most perceptually critical data, ensuring high-quality streaming even under adverse network conditions.
The TMP Model: Compressed-Domain Prioritization
The core innovation of the TMP model is its operation within the compressed domain. By analyzing metadata already present in the H.264 stream—such as motion vectors and macroblock (MB) partition sizes—the system avoids the high computational cost of full pixel-level decoding.
Priority for each Macroblock is determined by three key metrics:
- Texture Priority ($P_T$): Derived from MB partition sizes. Smaller partitions (e.g., 4x4) indicate high-frequency details that are difficult for decoders to "guess" or conceal if lost.
- Motion Priority ($P_M$): Distinguishes between predictable global motion (camera pans) and complex local motion. Local motion is prioritized as it is harder to hide through standard error concealment.
- Attention Priority ($P_V$): Aligns with the Human Visual System (HVS), which is naturally drawn to moving objects. High local motion areas are flagged as "attention areas."
Using Flexible Macroblock Ordering (FMO), the model groups macroblocks of similar priority into specific slices. This allows the network layer to implement a selective ARQ strategy, prioritizing the retransmission of high-priority slices over less critical background data.
Experimental Setup
The model was evaluated using three standard video sequences: Coastguard (complex motion/texture), Edberg (benchmark sequence), and Foreman (talking head with panning). Testing was conducted under a Bernoulli loss model with packet loss rates (PLR) ranging from 5% to 20% and bitrates between 500Kbps and 1200Kbps.
Results and Discussion
The proposed TMP-based selective ARQ method demonstrated superior performance across all test cases. Objectively, the system achieved an average gain of 1.07dB in PSNR compared to standard ARQ methods.
Subjective quality, measured via Weighted PSNR (WPSNR), showed even more significant improvements. By focusing protection on "attention areas," the model maintained visual clarity where it matters most to the viewer, even as network conditions worsened.
The system successfully maintained these quality gains while keeping end-to-end latency below 200ms, making it highly suitable for interactive applications like video conferencing.
Conclusion
The TMP model provides a lightweight, efficient solution for protecting video streams in error-prone environments. By leveraging compressed-domain analysis and selective ARQ, it achieves significant perceptual quality gains without the overhead of traditional content-aware methods. Future work will explore Hybrid ARQ (HARQ) schemes, combining this prioritization with Forward Error Correction (FEC) to further enhance resilience in high-latency networks.
Summary updated