a processor configured to execute machine-readable instructions to cause the device to crop a low resolution SEM image to generate a first cropped image and a second cropped image, the low resolution SEM image including an image of at least one of fins of a semiconductor device, shallow trench isolations of the semiconductor device, or gate lines of the semiconductor device,

to upscale the first cropped image and the second cropped image to generate a first upscaled image and a second upscaled image using a ground truth image having high-resolution and corresponding to the first and second cropped image, respectively, and

to cancel noise from the first upscaled image and the second upscaled image to generate a first noise canceled image and a second noise canceled image, wherein

the noise canceling includes canceling noise based on a deep learning network trained on a layout of the semiconductor device, the layout corresponding to a field of view of the SEM image.