LAUSR

The haze removal technique refers to the process of reconstructing haze-free images from scenes of inclement weather conditions. This task has an extensive demand in practical applications. At present, models based on deep convolution neural networks have made significant progress in the haze removal field, greatly outperforming the traditional prior and constraint methods. However, the current CNNs methods, which involve only a single input image, do not provide sufficient features to determine the optimal transmission maps for haze removal; therefore, we propose and design an aggregated resolution convolution network (ARCN) that uses multiple inputs and aggregates features from a CNN model and the adversarial loss algorithm. Experiments comparing the visual results of our network with those of several previous methods reveal substantial improvements.