Bayesian Neural Networks for One-to-Many Mapping in Image Enhancement

2025-01-24

AI Summary

To address the inherent one-to-many ambiguity in low-light and underwater image enhancement, this paper proposes the Bayesian Enhancement Model (BEM)—the first Bayesian modeling framework explicitly designed for one-to-many image enhancement mappings. BEM adopts a two-stage architecture that jointly leverages low-dimensional latent-space modeling and Bayesian neural networks (BNNs) to generate diverse, high-fidelity enhanced outputs. A novel Momentum Prior is introduced to dynamically regularize the BNN’s weight distribution, significantly accelerating training convergence and improving uncertainty calibration. Extensive experiments on multiple benchmarks demonstrate that BEM consistently outperforms deterministic methods in both fidelity metrics (PSNR, SSIM) and diversity measures, while maintaining real-time inference speed (>30 FPS). This work establishes a new paradigm for interpretable, controllable image enhancement under epistemic uncertainty.

Technology Category

• Machine Learning: Bayesian Learning
• Reasoning under Uncertainty: Relational Probabilistic Models
• Computer Vision: Diffusion Models for Vision

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

In image enhancement tasks, such as low-light and underwater image enhancement, a degraded image can correspond to multiple plausible target images due to dynamic photography conditions, such as variations in illumination. This naturally results in a one-to-many mapping challenge. To address this, we propose a Bayesian Enhancement Model (BEM) that incorporates Bayesian Neural Networks (BNNs) to capture data uncertainty and produce diverse outputs. To achieve real-time inference, we introduce a two-stage approach: Stage I employs a BNN to model the one-to-many mappings in the low-dimensional space, while Stage II refines fine-grained image details using a Deterministic Neural Network (DNN). To accelerate BNN training and convergence, we introduce a dynamic emph{Momentum Prior}. Extensive experiments on multiple low-light and underwater image enhancement benchmarks demonstrate the superiority of our method over deterministic models.

Problem

• Image Enhancement
• Multiple Solutions
• Uncertainty

Innovation

• Bayesian Enhanced Model (BEM)
• Dynamic Momentum Prior
• Bayesian Neural Networks (BNNs)