Low-Light Image Enhancement Model with Nonlinear Enhancement Probability

Because low-light image enhancement is of great significance for improving the accuracy of object detection, it has gained the attention of many researchers. In the process of low-light image enhancement, darker pixels have a greater probability of being enhanced, while brighter pixels have a greater probability of being suppressed. Therefore, based on the Zero DCE, we propose a low-light image enhancement model with nonlinear enhancement probability, named TransNEP. On the one hand, an enhancement function with attention is proposed to simultaneously enhance over-dark pixels and suppress over-bright pixels with a greater probability than the existing methods. On the other hand, we split the low-light image into multiple patches and embed these patches as patch queries. Then, based on the transformer’s key-query mechanism, patch queries are decoded to obtain the enhancement probability corresponding to each patch. Finally, this study verify the improvement in texture and object detection’s accuracy of the image enhanced with TransNEP through comprehensive experiments.