Abstract
Distributed video coding (DVC) is a paradigm that shifts computational complexity from the encoder to the decoder, making it ideal for resource-constrained devices like wireless sensors and mobile cameras. However, a significant performance gap remains between DVC and traditional codecs. This paper proposes a frame-based skip mode decision algorithm to bridge this gap. By utilizing an encoder-side Side Information (SI) quality estimation technique, the system accurately decides when to skip Wyner-Ziv (WZ) frames. Experimental results demonstrate a 50% reduction in decoding time and an average Rate-Distortion (R-D) improvement of 0.49 dB compared to the DISCOVER codec.
Introduction
Traditional video coding standards like H.264/AVC rely on complex encoders to achieve high compression efficiency. In contrast, Distributed Video Coding (DVC) targets applications where the encoding device has limited power. While DVC theoretically offers comparable performance by moving motion estimation to the decoder, practical implementations often suffer from a "performance gap."
A primary cause of this inefficiency is the lack of diverse coding modes. In particular, "skip modes"—which avoid transmitting data for unchanged regions—are often absent or overly complex in DVC. This paper introduces a low-complexity, frame-based skip mode decision to optimize both bitrate and processing time by identifying frames where the predicted Side Information is already of sufficient quality.
Proposed Methodology
The proposed system integrates a frame mode decision block at the encoder, as illustrated in the architecture below. This block determines whether a WZ frame should be processed normally or skipped entirely.
The core of the algorithm is the Encoder-side SI Quality Estimation. Since the actual Side Information is only available at the decoder, the encoder estimates its quality by comparing successive key frames. To align with the Human Visual System (HVS), a Gaussian filter is applied to the Mean Square Error (MSE) calculation, prioritizing the center of the frame where viewers typically focus.
A novel metric, $\vartheta$, is defined as the sum of the mean and standard deviation of the filtered MSE. This metric is designed to detect both localized high-motion objects and global camera panning. If $\vartheta$ falls below a predefined threshold, the entire WZ frame is skipped, saving the bits that would have been used for parity data.
Experimental Evaluation and Results
The algorithm was tested on standard QCIF sequences: Hall (static surveillance), Foreman (medium motion), and Coastguard (global motion/panning). The results were benchmarked against the well-known DISCOVER DVC codec.
As shown in the distortion variation plots, the $\vartheta$ metric (blue circles) closely tracks traditional quality measures like SSIM and MSE. In the Hall sequence, the consistently low $\vartheta$ values allowed the system to skip 62% of WZ frames. In the Coastguard sequence, the metric successfully captured the distortion spikes caused by camera panning, ensuring parity bits were sent only when necessary to maintain quality.
Key performance metrics include:
- Efficiency: Decoding time was reduced by approximately 50%, while encoding time saw a 9% reduction.
- Quality: The R-D performance improved by an average of 0.49 dB (BD-PSNR) over the DISCOVER benchmark.
- Low Overhead: The skip decision logic is highly efficient, adding negligible time to the encoding process.
Conclusion
By implementing a frame-level skip decision based on perceptual weighting and a robust quality metric, this research significantly enhances DVC efficiency. The approach effectively handles global motion and static backgrounds, providing substantial gains in decoding speed and video quality. While some edge cases involving sudden peripheral motion remain a challenge, the proposed method offers a practical path toward more competitive distributed video codecs.
Summary updated