The three pillars of our compression are image sensor characterisation and modelling, accurate noise replacement, and metrological tests.
‍
Most of the entropy of a given pixel value can be attributed to noise, namely about 9 bpp on a well-exposed 16 bpp sensor, and only about 1 bpp is actual information (signal). Signal and noise are mixed in a complex way, it’s impossible to deterministically distinguish them, unless one knows the signal.

Jetraw technology is based on ‘untangling’ information from noise by calibrating the sensor, thus enabling the high compression ratio. ‘Untangling’ cannot be done fully, as Jetraw is still bound by the rules of information theory. Reduction of signal-to-noise (SNR) is kept at a minimum by enforcing strictly bounded, uniform, unbiased and uncorrelated errors.