Low-light Image Enhancement via a Deep Hybrid Network

In this paper, we are proposing a trainable hybrid network to enhance the visibility of degraded images. Camera sensors often fail to capture clear images or videos in a poorly-lit environment which results in degraded images. 

Our proposed network consists of two distinct streams which simultaneously learn the global content and salient structures of the clear image in a unified network. More specifically, the content stream estimates the global content of the low-light input through an encoder-decoder network. 

However, the encoder in the content stream tends to lose some structure details. As a remedy , we proposed a novel spatially variant Recurrent Neural Network (RNN) as an edge stream to model edge details with the guidance of another auto-encoder. Experimental results shows the performance of proposed network favorably, against the state-of-the-art low-light image enhancement algorithms.