A novel method for video enhancement - RGB local context-based fusion

This paper presents a solution for image/video quality enhancement, with inputs being low quality nighttime images/videos and outputs being high quality ones. The developed approach here relies on context-based fusion, a group of solutions that need one more background image to support them, and thus, obtaining high quality results in comparison with other solutions. However, the major downside of these context-based fusion methods is too complex to apply into real-time applications, especially with high-resolution images/videos. Since they need much more hardware resources and take more time to run, thereby being hard to deploy them into dedicated hardware platforms. Due to this, our proposed method attempts to decrease the algorithm complexity in which the quality of the output images/videos is still kept. The experimental results show that our proposed method achieves both high-quality output images and time-effective results. More specifically, almost all of important information of the input nighttime images/videos is not only kept, but also enhanced, which is observed better; the vision of many unimportant dark areas of surrounds is seemly more natural; and aliasing, ghosting and haloing effects, which normally exist in previous studies, are likely to disappear. Another advantage is that the proposed method here outperforms both the Denighting method [5] and the Enhancing Dynamic Scenes in Gradient Field method [2], which is demonstrated by a comparison between processing times of these three methods with different resolutions.