Gated Fusion Network for Single Image Dehazing

摘要: In this paper, we propose an efficient algorithm to directly restore a clear image from hazy input. The proposed hinges on end-to-end trainable neural network that consists of encoder and decoder. is exploited capture the context derived input images, while decoder employed estimate contribution each final dehazed result using learned representations attributed encoder. constructed adopts novel fusion-based strategy which derives three inputs original by applying White Balance (WB), Contrast Enhancing (CE), Gamma Correction (GC). We compute pixel-wise confidence maps based appearance differences between these different blend information preserve regions with pleasant visibility. yielded gating important features inputs. To train network, introduce multi-scale approach such halo artifacts can be avoided. Extensive experimental results both synthetic real-world images demonstrate performs favorably against state-of-the-art algorithms.