Frequency and Spatial Dual Guidance for Image Dehazing

Abstract

Image dehazing is quite challenging in dense-haze scenarios, where quite less original information remains in the hazy image. Though previous methods have made marvelous progress, they still suffer from information loss in content and color in dense-haze scenarios. The recently emerged Denoising Diffusion Probabilistic Model (DDPM) exhibits strong generation ability, showing potential for solving this problem. However, DDPM fails to consider the physics property of dehazing task, limiting its information completion capacity. In this work, we propose DehazeDDPM: A DDPM-based and physics-aware image dehazing framework that applies to complex hazy scenarios. Specifically, DehazeDDPM works in two stages. The former stage physically models the dehazing task with the Atmospheric Scattering Model (ASM), pulling the distribution closer to the clear data and endowing DehazeDDPM with fog-aware ability. The latter stage exploits the strong generation ability of DDPM to compensate for the haze-induced huge information loss, by working in conjunction with the physical modelling. Extensive experiments demonstrate that our method attains state-of-theart performance on both synthetic and real-world hazy datasets.

FSDGNet: frequency and spatial dual guide network for crack detection

“…Phase conveys more information about the image structure than amplitude and is less affected by noise and contrast distortion 55 . Based on this theory, we find that the detected crack loss is mainly in the phase spectrum, while the amplitude spectrum is less affected.…”

“…Phase conveys more information about the image structure than amplitude and is less affected by noise and contrast distortion. 55 Based on this theory, we find that the detected crack loss is mainly in the phase spectrum, while the amplitude spectrum is less affected. This is also consistent with the conclusion that the light contrast of an image is represented by the amplitude spectrum and the texture structure information by the phase spectrum.…”

Multi-channel feature fusion attention Dehazing network

“…Figs 10 – 13 show the results of the ClarifyNet [ 40 ], FSDGN [ 41 ], MIRNetv2 [ 42 ], DehazeFormer [ 43 ], and UMFFA techniques on these two datasets, respectively. In all of the cases, the UMFFA technique produces the best outcomes.…”

UDC Image Restoration via Lightweight Dual-Domain Network

“…Although many approaches use frequency domain information to assist in image processing 26,27 , they do not consider the semantic information provided by phase and amplitude. Yu et al 28 reconstruct the phase under the guidance of amplitude, and obtain results with better texture details. Unfortunately, their reconstruction process is one-way, while we further use the phase information to feed back the enhancement of the amplitude.…”