Impact Analysis of Baseband Quantizer on Coding Efficiency for HDR Video

Chau-Wai Wong, Guan-Ming Su, Min Wu

TL;DR

This paper investigates how reducing the bit depth of High Dynamic Range (HDR) video signals before they enter a standard video codec (a process called baseband quantization) impacts compression efficiency. They analytically and experimentally demonstrate that this pre-compression step significantly reduces coding efficiency, suggesting that any necessary bit-depth reduction should occur within the video codec itself.

What Problem Does This Paper Address?

High Dynamic Range (HDR) video captures a much wider range of brightness and color detail than standard High Definition (HD) video, often requiring 10-12 bits per color channel compared to the 8 bits of traditional video. However, the vast majority of existing video codecs are designed for 8-bit or 10-bit video. To leverage these legacy codecs for HDR content, a common practice is to reduce the dynamic range of the HDR signal before encoding. This "baseband quantization" aims to fit the signal into the codec's capabilities. The crucial question this paper addresses is: does this pre-compression range reduction negatively impact the overall efficiency of compressing HDR video? In other words, is it better to reduce the signal's dynamic range before encoding, or let the codec handle it?

Methods

The authors develop a theoretical framework to analyze the impact of baseband quantization on coding efficiency. They model the video coding pipeline, including the baseband quantizer (Q1) and the codec quantizer (Q2), and derive mathematical expressions for the mean-squared error (MSE) and Signal-to-Noise Ratio (SNR) loss. A key step is simplifying the complex predictive coding loop to focus on the error introduced by quantization. They then use these theoretical models to predict the SNR loss as a function of the relative strengths of the baseband and codec quantizers. To validate their theory, they conduct simulations using synthetic data and real-world HDR video sequences, comparing the performance of codecs with and without significant baseband quantization.

Datasets

The paper uses three standard HDR video sequences: BALLOONFESTIVAL, MARKET, and TYPEWRITER. These sequences are provided in the BT.2020 color space and are encoded using the HEVC/H.265 standard. The authors also utilize synthetic data generated from AR(1) processes to verify their theoretical models. No new datasets are introduced.

Key Insights and Ideas

  • Baseband Quantization is Detrimental: The core finding is that reducing the bit depth of HDR video before it enters the codec actively harms compression efficiency. This is counter-intuitive, as one might expect fitting the signal into a smaller range to be beneficial.
  • Codec Quantization is Key: The paper suggests that if bit-depth reduction is necessary, it should be handled by the video codec itself. The codec's internal quantization mechanisms are designed to optimize for rate-distortion performance, and applying quantization before this process disrupts that optimization.
  • Theoretical Framework: The development of a tractable mathematical model to quantify the error introduced by both baseband and codec quantization is a significant contribution, allowing for a principled analysis of their combined effects.
  • Predictable Loss: The theoretical models accurately predict the SNR loss observed in simulations, providing strong evidence for the analysis. The loss is shown to be dependent on the relative strengths of the baseband and codec quantizers.

Key Results

  • Analytical Prediction: The theoretical analysis predicts that strengthening the baseband quantizer (i.e., reducing its bit depth) leads to a decrease in coding efficiency.
  • Experimental Validation: Simulations with synthetic data and real HDR video sequences confirm the theoretical predictions.
  • Quantifiable Loss: For a specific scenario where the baseband quantizer is strengthened by 1.6 bits (from an effective bitdepth of 9.8 to 8.2), the PSNR drop at high bitrates can be as much as 1.60 dB. This is a substantial loss in video quality for a given bitrate.
  • Independence of Block Size: The observed performance degradation is shown to be largely independent of block size and neighborhood correlation in the video data.

Limitations and Future Work

  • Linear Mapping Focus: The theoretical analysis primarily focuses on linear mapping for baseband quantization. The paper acknowledges that in practice, nonlinear mapping is often used for range reduction.
  • Non-uniform Quantization: The authors suggest that it would be interesting to investigate whether non-uniform baseband quantization also leads to a penalty in coding efficiency, as their current analysis is based on uniform quantization.
  • Perceptual Quality: While PSNR is used as a metric, the paper hints at the importance of perceptual quality, especially with HDR content. Future work could explore the perceptual impact of baseband quantization.

Summary updated

Abstract

Digitally acquired high dynamic range (HDR) video baseband signal can take 10-12 bits per color channel. It is economically important to be able to reuse the legacy 8 or 10-bit video codecs to efficiently compress the HDR video. Linear or nonlinear mapping on the intensity can be applied to the baseband signal to reduce the dynamic range before the signal is sent to the codec, and we refer to this range reduction step as a baseband quantization. We show analytically and verify using test sequences that the use of the baseband quantizer lowers the coding efficiency. Experiments show that as the baseband quantizer is strengthened by 1.6 bits, the drop of PSNR at a high bitrate is up to 1.60 dB. Our result suggests that in order to achieve high coding efficiency, information reduction of videos in terms of quantization error should be introduced in the video codec instead of on the baseband signal.
BibTeX
@article{wong2016impact,
  author = {Chau-Wai Wong and Guan-Ming Su and Min Wu},
  title = {{Impact Analysis of Baseband Quantizer on Coding Efficiency for HDR Video}},
  journal = {IEEE Signal Processing Letters},
  year = {2016},
  url = {https://openreview.net/forum?id=KdQ5Yc4jkWp},
}
Figure