Dual-Path Hyperprior Informed Deep Unfolding Network for Image Compressive Sensing

2026-05-10

AI Summary

This work addresses the limitations of existing deep unrolling networks in image compressive sensing, which typically process all image regions uniformly through a single measurement stream, thereby failing to exploit multi-source information or account for varying texture complexity. To overcome this, we propose a dual-path deep unrolling network guided by a hyperprior: the measurements are split into two subsets, and a lightweight hyperprior branch learns multi-domain priors to dynamically inform the reconstruction branch. This enables spatially adaptive step sizes and a gradient-aware hard/soft attention mechanism that selectively focuses on challenging-to-reconstruct regions. Experimental results demonstrate that the proposed method significantly outperforms state-of-the-art compressive sensing approaches, achieving notable improvements in both reconstruction accuracy and fine detail recovery.

Abstract

Recent Deep Unfolding Networks (DUNs) have significantly advanced Compressive Sensing (CS) by integrating iterative optimization with deep networks. However, existing DUNs still suffer from two challenges: 1) Reliance on a single measurement stream, which limits effective information interaction across distinct measurement subsets. 2) Uniform processing of all image regions, which overlooks varying reconstruction difficulties induced by diverse textures. To address these limitations, a novel Dual-Path Hyperprior Informed Deep Unfolding Network (DPH-DUN) is proposed, which partitions measurements into double subsets to enable hyperprior-guided reconstruction via a dual-path architecture. In the Deep Hyperprior Learning branch, a series of lightweight neural modules are designed to efficiently generate hyperprior knowledge of different domains, enabling collaborative guidance for the CS reconstruction. In the Hyperprior Informed Reconstruction branch, a deep unfolding framework with hyperprior guidance is constructed to iteratively refine reconstruction. Specifically, i) in the gradient descent step, a Hyperprior Informed Step Size Generation network is designed to dynamically generate spatially varying step maps, enabling adaptive fine-grained gradient updates. ii) In the proximal mapping step, two well-designed hyperprior informed attention mechanisms are introduced to dynamically focus on challenging regions via gradient-based hard and soft attentions, facilitating CS reconstruction accuracy. Extensive experiments demonstrate that the proposed DPH-DUN outperforms existing CS methods.

Problem

• Compressive Sensing
• Deep Unfolding Networks
• Measurement Partitioning
• Adaptive Reconstruction
• Texture-aware Processing

Innovation

• Dual-Path Architecture
• Hyperprior Guidance
• Deep Unfolding Network
• Adaptive Step Size
• Attention Mechanism