Why does MATLABs "imwrite" scale 12-bit images and how to circumvent this?

Question

I have an image given as an N x M - matrix with 12-bit data and I want to use imwrite to save the image as a .pgm file.

Why does MATLAB scale the image to 16bit? How can I circumvent this?

Using the 'MaxValue' argument also seems to alter the image, as it can't be displayed properly afterwards e.g. in IrfanView.

Answer 1

The MaxValue parameter is a little counter intuitive. It does specify to write the PGM tagged with a max value of a certain depth (e.g. 12-bit here), but it also tells iwwrite to rescale the data. The rescaling happens in writepnm>remap_pixel_values:

function newdata = remap_pixel_values(data, maxval)
%REMAP_PIXEL_VALUES Remap pixel values in array of pixel values.
%
%   NEWDATA = REMAP_PIXEL_VALUES(DATA, MAXVAL) remaps the pixel values in
%   DATA as follows
%
%   Class of DATA   Input                    Output
%   -------------   -----                    ------
%   uint8           The set {0,1,...,255}    The set {0,1,...,maxval}
%   uint16          The set {0,1,...,65535}  The set {0,1,...,maxval}
%   double          The interval [0,1]       The set {0,1,...,maxval}

So with uint16 data, it will rescale the data by applying a scale of 65535/maxval via the bit shift bitshift(data,-4);. You don't want it to rescale the data, but you also want it to write the file as 12-bit (this happens in writepnm>write_raw_data. The workaround is to apply the opposite scale before calling imwrite:

Iscaled = uint16(double(I)*(2^16-1)/(2^12-1))
imwrite(Iscaled,'test.pgm','MaxValue',2^12-1)

Note that you could use double values scaled between [0,1] too, according to the table in the above code comments.

For reading 12-bit PGM/PPM, see here.