We present a novel deep learning based approach-the Range Scaling Global U-Net (RSGUNet)-for perceptual image enhancement on mobile devices.
Perceptual image enhancement on mobile devices—smartphones in particular—has drawn increasing industrial efforts and academicinterests recently. Compared to digital single-lens reflex (DSLR)cameras, cameras on smart phones typically capture lower-quality imagesdue to various hardware constraints. Without additional information, itis a challenging task to enhance the perceptual quality of a single imageespecially when the computation has to be done on mobile devices. Inthis paper we present a novel deep learning based approach—the RangeScaling Global U-Net (RSGUNet)—for perceptual image enhancementon mobile devices. Besides the U-Net structure that exploits image featuresat different resolutions, proposed RSGUNet learns a global featurevector as well as a novel range scaling layer that alleviate artifacts inthe enhanced images. Extensive experiments show that the RSGUNetnot only outputs enhanced images with higher subjective and objectivequality, but also takes less inference time. Our proposal wins the1st place by a great margin in track B of the Perceptual Image Enhancementon Smartphones Challenge (PRIM2018).