Aiming at the problem of poor fusion quality of traditional algorithms, especially the lack of texture information in infrared images or visible images inability to obtain sufficiently bright images results in images with poor signal-to-noise ratios and superimposed significant read noise in lousy weather conditions, a deep learning method of infrared-visible images fusion based on encoder-decoder architecture is proposed. The image fusion problem is cleverly transformed into the issue of maintaining the structure and intensity ratio of the infrared-visible image. The corresponding loss function is designed to expand the weight difference between thermal target and background. In addition, a single image super-resolution reconstruction based on a regression network is introduced to tackle the traditional network mapping function not suitable for natural scenes. The forward generation and reverse regression models are considered to reduce the irrelevant function mapping space and approach the authentic scene data through double mapping constraints. Compared with other state-of-the-art approaches, our experimental results achieve appealing performance on visual effects and objective assessments. In addition, we can stably provide high-resolution reconstruction results consistent with the human visual observation in different scenes while avoiding the trade-off between spatial resolution and thermal radiation information typical of conventional fusion imaging.
|