CPC G06T 5/92 (2024.01) [G06T 3/4046 (2013.01); G06T 3/4053 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01)] | 20 Claims |
1. A hybrid method for low light image enhancement, the method comprising:
capturing, with an imager, a high resolution, low light image;
downsampling, with one or more processors operable with the imager, the high resolution, low light image to obtain a low resolution, low light image;
processing, by the one or more processors using a low light enhancement model of a deep neural network, the low resolution, low light image to obtain a low resolution, enhanced image; and
generating, by the one or more processors from a mathematical model, a high resolution, enhanced image from three inputs:
the high resolution, low light image;
the low resolution, low light image; and
the low resolution, enhanced image;
wherein:
the low light enhancement model is trained through deep learning of semantic enhancement for images photographed in a low light environment; and
the mathematical model comprises a sharpness-preserving mathematical model.
|