Multistage Progressive Single-Image Dehazing Network With Feature Physics Model

Deep learning has recently shown superior performance for single image dehazing. Most deep dehazing networks either estimate the parameters of atmospheric scattering model through convolutional neural network, or predict the hazy-free image without physical prior. These typical approaches ignore the dehazing mechanism in feature space. In addition, single encoder-decoder architecture is widely utilized to construct dehazing network, but it may suffer from limited semantic-level multi-scale contextual dependencies. To tackle these issues, we firstly propose a novel Physical model based Feature Enhancement Dehazing (PFED) block, which consists of a feature enhancement block and a feature dehazing block with two fully convolutional sub-networks. Such fully convolutional module can improve the adaptivity of non-uniform feature dehazing. On the basis of PFED block, we develop a Multi-Stage Progressive Dehazing Network (MSPD-Net), which progressively removes haze in a multi-stage architecture. In addition, the selective kernel feature fusion scheme is used to carry out cross-scale and cross-stage fusion, which can enable the model to capture the intra- and inter-stage interactions, respectively. Extensive experiments on three popular datasets demonstrate that MSPD-Net is comparable or even superior to the state-of-the-art methods. Specifically, MSPD-Net exceeds the Transformer based DehazeFormer-M 1.389dB on the SOTS-indoor dataset. Furthermore, a series of ablation experiments prove that the key components of our method can boost performance effectively.