A novel approach for enhancing very dark image sequences

• We propose a conceptual framework for the enhancement of very dark videos.
• We propose a denoiser boosting very small signal-to-noise ratio (SNR) video data.
• We propose a tone mapper enlarging the extremely low dynamic range of video images.
• We introduce a noise reduction scheme for normal-light videos with heavy noise.

We propose the methods, involving a denoiser and a tone mapper, that especially enhance very dark image sequences having an extremely small signal-to-noise ratio (SNR). The proposed denoiser and tone mapper, respectively, boosts the very small SNR and enlarges the extremely low dynamic range: they jointly form a conceptual framework to do the enhancement. We explicitly show that the conceptual framework adheres to a theoretical biological model of visual systems in nocturnal animals. The developed denoiser utilizes adaptive spatial–temporal filtering and effectively succeeds in reducing high noise and motion blur artifacts while preserving image details. The devised tone mapping method carries out different transformations for low and high intensity pixels by employing a single function which meanwhile takes into account the local contrast of each pixel, supporting a spatially and temporally consistent tone adjustment per frame. The conceptual framework and a customized post-process lead to a novel technique that is demonstrated to accomplish a satisfactory enhancement of very dark videos, and produce high quality restored image sequences – outperforming the recent state-of-the-art results. Finally, the proposed adaptive spatial–temporal filter and the post-processing procedure can be combined and adapted for normal-light image sequences to deal well with remarkably heavy noise.