Single Image Dehazing Using Hybrid Convolution Neural Network

Due to the dispersion of atmospheric particles, the quality of the image becomes inferior. The quality of the frames captured in poor weather conditions such as fog, smog, and haze declines as a result. These images are difficult to process using conventional visibility restoration algorithms. Several deep learning techniques have been proposed in order to resolve the aforementioned computer vision challenges. They are trained and tested on hazy dataset by designing and implementing a supervised learning-based convolutional neural network (CNN) or an unsupervised learning-based generative adversarial network (GAN). In this paper, a novel single image dehazing algorithm called hybrid CNN has been proposed that consists of wavelet and inverted wavelet-based multi-scale convolution neural network (WIW-MSCNN). Furthermore, the output of inverted wavelet transformed component has been processed using coarse scale network that generates an estimated transmission map; whereas, the output of the wavelet transformed component has been processed with the help of fine scale network along with the refinement of coarse transmission map. Images of various hazy datasets such as FRIDA, FRIDA2, RESIDE, NH-HAZE, O-HAZE, DENSE-HAZE, and HUDRS have been used for training and testing the architecture. The performance of the proposed model has been analyzed by comparing the generated results with existing dehazing techniques.