Single-image low-light enhancement via generating and fusing multiple sources

Imperfect lightness conditions usually lower the visual quality of an image by bringing in unclear image details and poor image contrast. Traditional low-light enhancement models based on one single input are often limited in avoiding the effect of over-enhancement or under-enhancement. Models based on fusing multiple input sources usually perform well in relieving this issue, as they can harmonize the complementary visual appearances of a same scene provided by different sources. Nevertheless, these models still have difficulty in dealing with the situation that only one input is at hand, which usually happens in many practical situations. In this paper, we propose a low-light enhancement model that artificially enriches input sources and then seamlessly fuses them. Specifically, with an input image, we first generate multiple enhanced images based on a lightness-aware camera response model. These images are then fused at mid-level based on a patch-based image decomposition model. To validate our model, we conduct qualitative and quantitative comparisons with several state-of-the-art single-source and multi-source models on a collection of real-world images. Experimental results show that our model better improves the image quality in terms of visual naturalness and aesthetics.