Semi‐supervised learning dehazing algorithm based on the OSV model

Despite the great progress that has been made in task of single image dehazing, results existing models restoring edge and texture information are still challenging. Besides, most dehazing trained on synthetic data, resulting poor generalization ability to real-world images. To address aforementioned problems, a semi-supervised learning method based decomposition model Osher, Solé, Vese(The OSV model) is presented. Specifically, first applied decompose hazy into structure layer layer, save dehaze for restore images with sharper edge. Furthermore, network adopts algorithm generative adversarial networks (GAN) generalize better images, which includes two branches: supervised unsupervised learning. Extensive experiments indicate proposed preserves more accurately while better, performs favourably against advanced algorithms both outdoor datasets