[yorak]

There is also superresolution methods that in practice make the "enhance"-feature possible. The information from the same objects in successive frames, eg. from a security camera feed, are composited into a single frame of higher resolution than the originals. This can even be done for a single image (see. Glasner et al. "Super-resolution from a single image." Computer Vision, 2009 ICCV 2009. IEEE.).

If machine learning is brought into play it gets even more interesting. See eg. waifu2x, the anime video upscaler based on Deep Convolutional Neural Networks. Basically it is a neural net, that has been shown A LOT of anime, and it can therefore creatively(?) fill in the created pixels when upscaling.

I have no idea how advanced this kind of ML boosted "enhance" technology can get in the future, and it might even get a little scary: imagine an observer with the experience of million lives spent looking blurry images in poor lighting. It might get surprisingly good at that...