Prior‐guided multiscale network for single‐image dehazing

Abstract

Single-image dehazing is an important problem because it is a key prerequisite for most high-level computer vision tasks. Traditional prior-based methods adopt priors generated from clear images to restrain the atmospheric scattering model and then recover haze-free images. However, these prior-based methods always encounter over-enhancement, such as halos and colour distortion. To solve this problem, many works use a convolutional neural network to retrieve original images. However, without priors as guidance, these learning-based methods dehaze effectively in synthetic datasets but perform poorly in real scenes. Hence, in this paper, we propose a prior-guided multiscale network for singleimage dehazing named PGMNet. Specifically, prior-based methods are adopted to acquire dehazed images of the training dataset in advance and then send these dehazed images to a parameter-shared encoder to form multiscale features. During the decoding process, these multiscale features are adopted to guide the prior-guided multiscale network to recover more image details. Moreover, considering that these prior-based dehazed images usually contain some over-enhanced regions, a spatial attention guided feature aggregation module and squeeze-and-excitation module are adopted to alleviate colour distortion. The proposed PGMNet takes the advantage of prior-based methods in real haze removal and provides superior performance compared with the state-of-the-art methods on both synthetic and real-world datasets.

Prior-combined dehazing network based on mutual learning

“…Hence, DANet [19] builds a bidirectional translation network to compensate the domain differences between synthetic images and real-world images. Moreover, PGMNet [20] extracts the features of hazy images and prior dehazed images by a siamese like encoder, and then fuses these features to achieve dehazing. Differently, PSD [21] proposes a multiple priors combined paradigm, which enhances the dehazing effect in real scenes but causes some illumination distortions.…”

“…In other worlds, the synthetic hazy images cannot represent uneven haze distribution and complex illumination in natural conditions, and make the trained model cannot hold in these scenes. Hence, some more recent works [19] [20] [21] [22] combine traditional priors (i.e. dark channel prior) with learning-based methods to achieve better dehazing effect in both synthetic and real scenes.…”

SSKDN: a semisupervised knowledge distillation network for single image dehazing

“…To this end, model-free methods 10 – 16 build end-to-end dehazing networks, and directly acquire dehazed images by learning the feature mapping between clear images and hazy images. Compared with model-based methods, these model-free methods can acquire dehazed images with better color fidelity.…”