A Two-Stage Dehazing Framework Based on Inverted Image Curve-Enhancement

Due to relying on accurate estimates of the atmospheric scattering parameters for atmospheric scattering models(ASM), existing haze removal methods suffer from some drawbacks, such as imbalanced contrast and missing details. To address the above issues, this paper proposes a two-stage dehazing network called CEDhazeNet. It consists of a contrast enhancement module (CE) and a texture detail restoration module (TDR). In CE, we introduce a novel curve enhancement dehazing model to tackle the problem of contrast imbalance. Specifically, we observe that inverting the hazy image can accentuate all the regions obscured by haze, and by enhancing the exposure of the inverted image, we can significantly reduce the impact of haze on image contrast. In TDR, we construct a multi-scale information distillation network, which use more effective information, such as edges and spots to recover the texture details of hazy image. Extensive experiments on both synthetic and real world hazy image datasets demonstrate that CEDhazeNet outperforms state-of-the-art haze removal methods in terms of quantitative accuracy and subjective visual quality.