Biologically-inspired two-pathway transformer network for image dehazing

Image dehazing, a classical low-level task in computer vision, is challenging due to the need to reproduce clean, haze-free information from variable haze conditions — making it an ill-defined problem. While convolutional neural networks (CNNs) have been widely used as learning-based tools in image dehazing, they possess inherent characteristics that limit their ability to effectively comprehend the interconnections among image information. This often leads to unnecessary redundancies and limited performance in CNN-based dehazing networks. Moreover, existing prior-based methodologies for image dehazing frequently fail due to the constraints of their statistical priors, which may be inconsistent with complex and changing environments. Inspired by biological vision processing mechanisms, we propose a two-pathway network to address these issues. Specifically, we design a decomposition network to decompose the input image into high-frequency and low-frequency components. Using a divide-and-conquer strategy, these two components are fed into high-frequency and low-frequency networks for individual detail restoration. A lightweight CNN is constructed for the low-frequency pathway to initiate the recovery of low-frequency details. For the high-frequency pathway, we design a network with multi-scale Transformer blocks to capture long-range dependencies within the feature space. Following this, the final enhanced image is generated by directly combining the two restored components. Through extensive experiments on large-scale image dehazing datasets, the proposed method exhibits significant improvements over state-of-the-art techniques in both qualitative and quantitative evaluation. © 2025 Elsevier Ltd.