Hybrid gated recurrent unit and convolutional neural network-based deep learning architecture-based visibility improvement scheme for improving fog-degraded images

The color and contrast of the captured images under weather conditions is considered to be degraded due to the airlight and attenuation of the radiance introduced by a scene visualized by the observer. At this juncture, a significant fog removal approach that attributes towards significant visibility improvement in order to reduce road accidents in turbid weather. Moreover, fog image dehazing visibility improvement techniques are considered to play an anchor role in minimizing the contrast that has a predominant impact over the driver assistance system (DAS) in the real world. In this paper, hybrid gated recurrent unit and convolutional neural network-based deep learning architecture (HGRU-CNN-DLA)-based visibility improvement scheme for improving fog-degraded images. The degree of visibility enhancement is considerably enhanced by the proposed HGRU-CNN-DLA by extracting time sequence data feature vector derived through GRU and other high dimensional feature vector achieved using CNN. In specific, GRU module is included for modeling dynamic transition in the past load sequence data for learning the superior features inherent with them. The module of CNN is particularly used for processing the matrices of spatio-temporal data and transform them into potential feature vector that aids in significant visibility restoration. The simulation experiments of the proposed HGRU-CNN-DLA is conducted to estimate its significance over the state-of-art CNN-based image visibility enhancement schemes evaluated using mean opinion score (MOS) and fog reduction factor (FRF).