研究目的
To propose a simple but powerful color attenuation prior for haze removal from a single input hazy image.
研究成果
The proposed approach achieves dramatically high efficiency and outstanding dehazing effects. However, a more flexible model is highly desired to overcome the challenge of the scattering coefficient β not being a constant in inhomogeneous atmosphere conditions.
研究不足
The scattering coefficient β in the atmospheric scattering model cannot be regarded as a constant in inhomogeneous atmosphere conditions.
1：Experimental Design and Method Selection:
The study proposes a linear model for modeling the scene depth of the hazy image under a novel color attenuation prior and learns the parameters of the model with a supervised learning method.
2：Sample Selection and Data Sources:
500 haze-free images are used for generating the training samples (500 random depth maps and 500 synthetic hazy images).
3：List of Experimental Equipment and Materials:
Not explicitly mentioned.
4：Experimental Procedures and Operational Workflow:
The depth information is recovered with the proposed color attenuation prior. With the depth map of the hazy image, the transmission is estimated and the scene radiance is restored via the atmospheric scattering model.
5：Data Analysis Methods:
The study uses a supervised learning method to determine the coefficients of the linear model and the variable σ2.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
