【摘要】 低光照图像/视频质量增强是一个普遍且具有挑战性的任务，在安防、医学成像及自动驾驶等很多领域都有重要的应用。为了得到更好的质量评价指标，大多数用于该任务的模型较为复杂，对其部署与运行来说具有挑战性。对此，设计了一种用于低光照图像/视频质量增强的方法。方法基于神经网络实现，网络仅166K的参数量，可实现全高清视频的实时处理。具体来说，网络以一个简化的3层编解码结构作为主体，为了增强网络性能，设计了一个高效的模块结构，模块去除了网络中常用的激活层，用加法和乘法的线性组合实现了激活函数的非线性功能。对比实验结果表明上述方法在低光照图像和视频数据集上都具有较好的性能，在保证低光照图像/视频纹理细节的同时有效提升亮度。更多还原

【Abstract】 Low-light image/video quality enhancement is a common and challenging task with significant applications in fields such as security,medical imaging,and autonomous driving. To achieve better quality metrics,most models used for this task are quite complex,making their deployment and operation challenging. In response,this paper designs a method for enhancing the quality of low-light images/videos. The method is implemented based on a neural network with only 166K parameters,enabling real-time processing of full HD videos. Specifically,the network adopts a simplified 3-layer encoder-decoder structure as its core. To enhance network performance,an efficient module structure was designed,removing the commonly used activation layers in the network and implementing the nonlinear functionality of activation functions through a linear combination of addition and multiplication. Comparative experimental results show that our method performs well on low-light image and video datasets,effectively improving brightness while preserving the texture details of low-light images/videos.更多还原