Progressive Update Guided Interdependent Networks for Single Image Dehazing

Images with haze of different varieties often pose a significant challenge to dehazing. Therefore, guidance by estimates of haze parameters related to its variety would be beneficial and they should be progressively updated along with iterative haze reduction to allow optimal dehazing. To this end, we propose a multi-network dehazing framework containing novel interdependent dehazing and haze parameter updater networks that operate within a unique iterative mechanism. The haze parameters, transmission map and atmospheric light, are first estimated using specific convolutional networks allowing color cast handling. The estimated parameters are then used as priors in our dehazing module, where the estimates are progressively updated by novel convolutional networks using the iterative mechanism. The updating takes place jointly with progressive dehazing by a convolutional network that invokes inter-iteration dependencies. The joint updating and dehazing within the iterative mechanism gradually modify the haze parameter estimates toward achieving optimal dehazing. Through ablation studies, our iterative dehazing framework is shown to be more effective than the use of conventional LSTM based recurrence, image-to-image mapping and haze model based estimation. Our dehazing framework is qualitatively and quantitatively found to outperform the state-of-the-art on synthetic and real-world hazy images of several datasets with varied hazy conditions.