[Blog Post] MSI at last!

[Bernd]

Those images are 16-bits-per-pixel, but because of the light levels and/or the camera sensitivity the values in each image range from 0 to some "maximal" limit between 1000 and 4000 or so.  Thus, the brightest pixels are less than 4000/65535 = 6% of "white".  Which looks like black when viewed with any software designed for "consumer" images.  Yet those images have at least 10 times as many gray levels as  a common image.

I understand that - but isn't this a huge waste? 94% of the available tonal range just is not used. It means the image is underexposed about 5 stops and this sensor headroom is wasted. Why not expose the image in a way that the histogram spans most of the range 0-65535? There's the spectralon target which has much higher albedo than the manuscript so there's no danger of blowing out highlights.

Also this is a still image. I understand that within a narrow wavelength band you might not have enough illumination with a moving target where exposure time is limited. But this is not the case here. If I have to push an image ~8 stops to get a decent exposure, then something's not right. And this both induces noise in mid-tones and shadows and reduces dynamic range. This has nothing to do with 'consumer's software. No software can recover the signal-noise ratio of a decently exposed image.


So why? Conservatory issues (exposing the VM to the minimum of light necessary)?But this is negated by the amount of images taken under different angles and wavelengths. I assume whoever made these images knew what he was doing but one should never assume. I learned that the hard way more than once. I still think we could pull significantly more information from a properly exposed UV image with better SNR and dynamic range.