The image has been cropped, but the assumption is that the number of pixels occupied by the image of the gyro remains the same (which is exactly the case for a simple crop, without additional interpolation).
As we are dealing with ratios (in this case the ratio of the number of pixels the gyro takes up compared to the number of pixels across the camera sensor) the arithmetic remains valid.
If the image had been manipulated to crop and interpolate, increasing the number of apparent pixels in the cropped image, then there should be clear indications of artifacts, that don't seem to be present in that image.
Although it's quite possible to retain the original image meta data in the image after a major Photoshop job, I think it's unlikely that this is the case here. The image records the camera as a Canon EOS40D, with a 3888 x 2592 pixel sensor, fitted with a 400mm (35mm equivalent) focal length lens. The image was later cropped to 1063 x 768 pixels. The cropped part is just a subset of the full sensor image, hence the reason that the ratiometric technique still works.
VP