Adaptive Image Defogging Algorithm Based on DCNN
Wuyunbilige, Licai Liu, Rui Wang, Shuai Gao, Ze Mao, Xiaohu Tang · 2020
Under the influence of special weather conditions such as heavy fog, haze, sand and dust, the images taken outdoors are gray and white due to the reflection of light by cluttered particles in the air. Most existing image defogging algorithms are single-layer network feature extraction, feature information is seriously lost, and transmittance calculation is inaccurate. In view of the above problems, an adaptive image defogging algorithm based on deep convolutional neural network is proposed. The implementation of this algorithm is still based on the atmospheric scattering model. There are three kinds of fully convolutional networks: shallow extraction, parallel extraction and deep fusion, which extract the shallow features of the image, extract the deep features and fuse the deep and shallow features together to make the transmission image more accurate. After experimental testing, compared with the traditional defogging algorithm, this deep convolution defogging algorithm has a better defogging effect on outdoor outdoor fog images, especially the defogging effect of details.