Feature Aggregation Convolution Network for Haze Removal
Linyuan He, Junqiang Bai, Meng Yang · 2019
The Haze Removal technique refers to the process of reconstructing haze free images from inclement weather conditions in the same scene, which has a more extensive demand in practical application. At present, the model based on the deep convolution neural network has made significant progress in the field of haze removal, and its effect has greatly exceeded the traditional prior and constraint methods. In view of the current CNN based dehazing methiods only take one input image into consideration so that they cannot capture enough features for indicating the optimal transmission maps, we propose and design a Feature Aggregation Convolution Network (FACN), with the multi inputs and feature aggregated of CNN model and adversarial loss algorithm. A comparative experiment with a few previous methods shows improvement visual results.