CIASM-Net: A Novel Convolutional Neural Network for Dehazing Image

Abstract

When light propagates in the medium such as haze, the image information collected by the imaging sensor is seriously degraded due to the scattering of particles, which greatly limits the application value of the image. In this paper, a novel convolutional neural network model called CIASM-Net is proposed to implement image dehazing. CIASM-Net includes color feature extraction convolutional networks and deep defogging convolutional networks. The color feature extraction convolution network is used to extract the color features of foggy images; the deep dehazing convolution network improves the inverse atmospheric scattering model convolution network, and uses a multi-scale convolution layer instead of the original convolution layer to estimate the transmittance value. Moreover, we add a pyramid pooling layer to the network to extract global features. To obtain the optimized network model, we use the classic RESIDE training set to train the network model. We have performed extensive experiments on the synthetic hazy dataset and the real-world hazy dataset in the RESIDE test set. The experimental results have proved that the model has satisfactory results.