CogSENet: Blind Image Deblurring with Blur-Conditioned Semantic Routing and Explicit Frequency Fusion

2026-06-29

AI Summary

Existing blind image deblurring methods struggle with spatially varying blur in real-world scenarios and lack semantic awareness to distinguish genuine textures from artifacts. Inspired by eagle-eye vision, this work proposes a dynamic semantic alignment framework that models long-range dependencies through a semantics-driven state space module and achieves physically interpretable structure and texture recovery via explicit wavelet-domain decomposition into high- and low-frequency components. The approach further incorporates a differentiable semantic routing mechanism, a bi-frequency fusion block (BFFB), and an adaptive modulation strategy that integrates CLIP-derived semantic priors with a continuous blur field. The resulting model surpasses current state-of-the-art methods in both visual quality and structural fidelity while using fewer parameters, and demonstrates strong generalization across image dehazing, deraining, and denoising tasks.

Abstract

Blind image deblurring demands the recovery of high-fidelity details and coherent structures from complex, unknown degradations. Current blind image deblurring methods struggle with real-world, spatially varying degradations, and lack the semantic awareness necessary to reliably differentiate valid textures from artifacts. To bridge this gap, we propose CogSENet, a dynamic, semantic-aligned reconstruction framework inspired by the eagle's visual system. By mimicking the eagle's active saccadic scanning, we devise a Semantic-Driven State Space Module (SDSSM) with semantic-aware token regrouping via differentiable routing, enabling prompt-conditioned long-range dependency modeling. To ensure physically interpretable recovery of textures and structures, a BiFreqFusionBlock (BFFB) mirrors functional differentiation of the eagle's retina by decomposing features into high and low frequencies using wavelet transforms. Finally, we estimate a continuous Blur Field (CBF) from blur image and fuse it with CLIP semantic priors to modulate the deepest latent features, emulating focal adaptation and enabling adaptive restoration under spatially non-uniform blur. Extensive experiments demonstrate that CogSENetoutperforms state-of-the-art deblurring methods in both visual quality and structural fidelity with fewer parameters, while also performing favorably on dehazing, deraining, and denoising tasks.

Problem

• blind image deblurring
• spatially varying degradations
• semantic awareness
• texture-artifact disambiguation
• real-world blur

Innovation

• Semantic-Driven State Space Module
• BiFreqFusionBlock
• Continuous Blur Field
• Wavelet-based Frequency Decomposition
• CLIP Semantic Prior