Encoder and Decoder-Based Feature Fusion Network for Single Image Dehazing

Single image defogging that aims to restore a fog-free image from its appropriately unconstrained hazy environment is a fundamental yet complex work that has recently achieved enormous interest. However, images reconstructed by certain available haze-removal approaches frequently retain artefacts, and color distortions, drastically degrading the visual quality and adversely affecting vision tasks. To that aim, we propose an encoder-decoder model that combines feature fusion with channel and color attention to improve real-time dehazing performance. Feature fusion block analyzes distinct features and pixels unequally, allowing for greater mobility in handling multiple types of input features and increasing model efficiency. The detailed quantitative and qualitative evaluation findings show that the suggested technique outperforms state-of-the-art techniques on dehazing data sets and real-time hazy images.