How can I save a 32 bit image (uint32) as a tiff?

Hello,

I am currently working on a post processing HDR script which requires that I use a 32 bit data type. The images start at 14 bit, once they are imported I convert them to int32, and after I do my calculations I convert them back to uint32. After I create my HDR image the max pixel value is about 181,000 counts. I am currently using,

imwrite(FinalImage, 'HDR Result.tif')

which works for 16 bit images, but does not seem to work for 32 bit images. If you could give me any help or insight I would greatly appreciate it.

Thank you, Jason

[Ashish Uthama]

Could you say more on what you mean by 'does not seem to work'? Do you get an error message or does it write the data incorrectly?

Answer 1

[DGM]

Here's a simple demo that should demonstrate writing odd numeric classes to a TIFF. This supports 8,16, and 32 bit signed and unsigned integers, as well as single and double-precision float. Bear in mind that many things in IPT and base MATLAB may not properly deal with int8, int32, or uint32 images, so you're on your own.

% an uint8 image
inpict = imread('peppers.png'); % uint8, RGB
% inpict = im2gray(inpict); % if you want to play with signed-int types, RGB is not supported

% make it into a 32b integer class somehow
% normally, i'd just use MIMT imcast(), but let's just do this garbage
%inpict = imcast(inpict,'uint32'); % let's pretend we can't have nice things
inpict = uint32(inpict)*16843009; % uint8 to uint32
%inpict = int32(double(inpict)*16843009 - 2147483648); % uint8 to int32

% output file name
filename = 'test.tiff';

% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% get to work
sz = size(inpict,1:3);
inclass = class(inpict); 
t = Tiff(filename, 'w');
tagstruct.ImageLength = sz(1);
tagstruct.ImageWidth = sz(2);
tagstruct.Compression = Tiff.Compression.AdobeDeflate;
tagstruct.SamplesPerPixel = sz(3);
tagstruct.PlanarConfiguration = Tiff.PlanarConfiguration.Chunky;

% I/RGB?
% i'm choosing to presume that 3ch inputs are RGB
if sz(3) == 1
    tagstruct.Photometric = Tiff.Photometric.MinIsBlack;
elseif sz(3) == 3
    tagstruct.Photometric = Tiff.Photometric.RGB;
end

% what is the numeric type?
if isinteger(inpict) && inclass(1) == 'u'
    tagstruct.SampleFormat = Tiff.SampleFormat.UInt;
elseif isinteger(inpict) && inclass(1) == 'i'
    tagstruct.SampleFormat = Tiff.SampleFormat.Int;
elseif isfloat(inpict)
    tagstruct.SampleFormat = Tiff.SampleFormat.IEEEFP;
else
    error('this demo only supports integer and float inputs')
end

% what's the sample width?
switch inclass
    case {'uint8','int8'}
        tagstruct.BitsPerSample = 8;
    case {'uint16','int16'}
        tagstruct.BitsPerSample = 16;
    case {'uint32','int32','single'}
        tagstruct.BitsPerSample = 32;
    case {'uint64','int64'}
        error('LibTIFF does not support 64b outputs except for double-precision floats')
    case 'double'
        tagstruct.BitsPerSample = 64;
    otherwise
        error('this demo only supports integer and float inputs')
end

% write everything and close the file
t.setTag(tagstruct);
t.write(inpict);
t.close();

% read the image back
newpict = imread(filename);
isequal(inpict,newpict) % test that it's the same

ans = logical
   1

Limitations:

• I have made no attempt to support logical images
• AFAIK, uint64/int64 inputs are not supported by LibTIFF. There may be broader reasons for that.
• Signed integer inputs cannot be RGB.