I would like to ask about the linearization in RAW processing. What does this term "linearization" mean in the theory of RAW-processing ?
The sensor output of image sensors are not always linear in nature. To make this non-linear response linear in RAW images(which contain the direct image sensor output), before demosaicing them to RGB from Bayer format(assuming the sensor outputs Bayer CFA images) and any other processing like white-balance adjustment, black level correction etc. Linearization block is used.
In case responses are not linearized the other processing blocks will introduce error as they assume the input data to be linear in nature. For example let's take the white-balance adjustment block depicted below pictorially :
just google first entry: (http://www.rawdigger.com/node/283)
"There are three disctinct linearization steps.
First one is applying tone decompression, and it is usually performed according to the compression table that is contained in the metadata that accompanies the raw file. Sometimes additional experiments are needed in order to determine what compression is in fact used, as the metadata is missing or not decyphered.
Second step (may be in the same table as the above) is lineraization of the extreme highlights.
The third one is to compensate for the non-linearity in shadows caused by the flare. To perform this step a raw converter must usually rely on user-supplied calibration data, because this type of linearization is strongly dependent on the particular lens sample, and also on some minute details such as the dust in the chamber. In the consumer world, I do not know of any commercial converter that cares to perform this compensation."