GGADN: Guided generative adversarial dehazing network

Image dehazing has always been a challenging topic in image processing. The development of deep learning methods, especially the generative adversarial networks (GAN), provides new way for dehazing. In recent years, many methods based on GAN have applied to However, two problems Firstly, For haze image, not only reduces quality but also blurs details image. network, it is difficult generator restore whole while removing haze. Secondly, model defined as minimax problem, which weakens loss function. It distinguish whether making progress training process. Therefore, we propose guided network (GGADN). Different from other generation networks, GGADN adds module generator. verifies each layer At same time, map generated by are strengthened. Network pre-trained VGG feature and L1-regularized gradient prior developed function parameters. From results synthetic images real images, proposed method better than state-of-the-art methods.