Progressive Update Guided Interdependent Networks for Single Image Dehazing

TL;DR

A multi-network dehazing framework containing novel interdependent dehazed and haze parameter updater networks that operate in a progressive manner is proposed and is qualitatively and quantitatively found to outperform the state-of-the-art on synthetic and real-world hazy images of multiple datasets with varied haze conditions.

Abstract

Images with haze of different varieties often pose a significant challenge to dehazing. Therefore, guidance by estimates of haze parameters related to the variety would be beneficial, and their progressive update jointly with haze reduction will allow effective dehazing. To this end, we propose a multi-network dehazing framework containing novel interdependent dehazing and haze parameter updater networks that operate in a progressive manner. The haze parameters, transmission map and atmospheric light, are first estimated using dedicated convolutional networks that allow color-cast handling. The estimated parameters are then used to guide our dehazing module, where the estimates are progressively updated by novel convolutional networks. The updating takes place jointly with progressive dehazing using a network that invokes inter-step dependencies. The joint progressive updating and dehazing gradually modify the haze parameter values toward achieving effective dehazing. Through different studies, our dehazing framework is shown to be more effective than image-to-image mapping and predefined haze formation model based dehazing. The framework is also found capable of handling a wide variety of hazy conditions wtih different types and amounts of haze and color casts. Our dehazing framework is qualitatively and quantitatively found to outperform the state-of-the-art on synthetic and real-world hazy images of multiple datasets with varied haze conditions.

Figures (40)

• Figure 1: A schematic diagram of our proposed multi-network framework 'PUG-D' for dehazing. (Dotted red lines indicate progressively updated content)
• Figure 1: (c) & (e) Hazy images generated from the (a) haze-free outdoor images applying the Koschmieder's model using the (b) & (d) depth maps, respectively. Depth maps (RESIDE) are as given in the RESIDE dataset and Depth maps (Ours) are generated using li2018megadepth for our NR-haze dataset. [Depth map grading: darker is nearer, brighter is farther]
• Figure 2: Subjective comparison of our result on a real hazy image peng2019image with the state-of-the-art MSBDN Dong_2020_CVPR and D4 yang2022self methods for dehazing. Cropped regions in boxes are for detailed inspection.
• Figure 2: Sample hazy images from the NR-haze dataset. The first row images, except the rightmost, are from LSIT, MSIT and HSIT that represent indoor images with low, mid and high haze, respectively. The second row images, except the rightmost, are from LSOT, MSOT and HSOT that represent outdoor images with low, mid and high haze, respectively. The color cast hazy images in the last column are taken from SCHT.
• Figure 3: Atmospheric light estimation model to predict atmospheric light across $3$ color channels. The convolutional blocks hierarchically extract regional contributions to atmospheric light, which is pooled globally to get the maximum contribution as the estimate.