Conditional Rate-Distortion-Perception Trade-Off

Xueyan Niu,Deniz Gündüz,Bo Bai,Wei Han

2023-05-22

Abstract:Recent advances in machine learning-aided lossy compression are incorporating perceptual fidelity into the rate-distortion theory. In this paper, we study the rate-distortion-perception trade-off when the perceptual quality is measured by the total variation distance between the empirical and product distributions of the discrete memoryless source and its reconstruction. We consider the general setting, where two types of resources are available at both the encoder and decoder: a common side information sequence, correlated with the source sequence, and common randomness. We consider both the strong perceptual constraint and the weaker empirical perceptual constraint. The required communication rate for achieving the distortion and empirical perceptual constraint is the minimum conditional mutual information, and similar result holds for strong perceptual constraint when sufficient common randomness is provided and the output along with the side information is constraint to an independent and identically distributed sequence.

Information Theory

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: in lossy compression, how to balance the relationship among transmission rate, distortion and perception quality. Specifically, the authors study the trade - off among these three when the perception quality is measured by the total variation distance between the empirical distribution and the joint distribution of the discrete memoryless source and its reconstruction.

### Problem Background
The traditional rate - distortion theory mainly focuses on how to achieve the target distortion level with the minimum transmission rate, but this theory does not fully consider the quality of human perception. Recent research shows that traditional distortion measures such as mean - squared error cannot well reflect the perception quality of reconstructed data. Therefore, the introduction of perception constraints becomes particularly important.

### Main Contributions
1. **Introduction of Perception Constraints**: The paper introduces two types of perception constraints - empirical perceptual constraint and strong perceptual constraint. The former only cares about the similarity of empirical distributions, while the latter also requires the consistency of sample order.
2. **Resource Utilization**: The situation where the encoder and decoder simultaneously have common side information and common randomness is studied. These resources can help better maintain the perception quality.
3. **Theoretical Framework**: The conditional rate - distortion - perception function is proposed, and the corresponding rate regions are derived. For the strong perceptual constraint, specific formulas are given when there is sufficient common randomness and the output sequence and side information are independent and identically distributed.
4. **Achievability and Converse Proofs**: Through strict mathematical proofs, it is shown how to achieve specific rate - distortion - perception combinations under given conditions, and the optimality of these results is proved.

### Formula Summary
- **Empirical Perceptual Constraint**:
\[
R^{(e)}(\Delta, \Pi)=\inf_{P_{Y|XZ}: E[D(X,Y)]\leq\Delta, d_{TV}(P_X, P_Y)\leq\Pi}I(X;Y|Z)
\]
- **Strong Perceptual Constraint**:
\[
R^{(s)}(\Delta, \Pi)\leq\inf_{P_{Y|XZ}: E[D(X,Y)]\leq\Delta, \limsup_{n\rightarrow\infty}d_{TV}\left(\prod_{i = 1}^n P_X, \prod_{i = 1}^n P_Y\right)\leq\Pi}I(X;Y|Z)
\]

Through these studies, the paper provides a theoretical basis and method guidance for how to better combine perception quality in lossy compression in the future.