A unified weighted variational model for simultaneously haze removal and noise suppression of hazy images

Visibility degradation of outdoor hazy images caused by scattering medium (e.g. haze, fog or smoke) is a challenging issue due to its ill-posed nature. Moreover, the noises are inevitably introduced to the hazy images because of various factors in real-world haze imaging. However, traditional atmospheric scattering model (ASM) fails to consider the noise term, which may lead to noise amplification after restoration. In this paper, traditional ASM is first modified by adding the noise term to make it suitable for real-world hazy images. Based on the modified model, a novel unified weighted variational model is proposed to address the problems of both haze removal and noise suppression simultaneously. Specifically, we use ℓ1 norm to enforce the piece-wise continuous on the scene radiance, adopt weighted ℓ2 norm based on exponentiated mean local variance to constrain the transmission, and apply the overall intensity constraint on the noise. An alternating optimization method is adopted to solve the proposed model. Compared with various dehazing methods, the proposed method can provide comparable or better results both qualitative and quantitative assessments on real-world and synthetic hazy images.