How can I write 32-bit floating point TIFs with NaNs?

I'm able to use IMREAD to read 32-bit floating point TIFs with NaNs for dead pixels.

After this, I do an image rotation with IMTRANSFORM with 'FillValue' of NaN (IMROTATE only fills with zeros).

Now, I'd like to save 32-bit floating point TIFs with NaNs in tact for later.

How can I do this? I've tried IMWRITE and MULTIBANDWRITE, but they don't seem to work.

I'm using MATLAB 7.11.0 (R2010b).

Thanks,

Kevin Crosby

Answer 1

[Walter Roberson]

I suggest using the Tiff class and specifying RGB for the Photometric and specifying 32 bits per channel and writing single precision values.

http://www.mathworks.com/help/matlab/ref/tiff-class.html

This is based upon the assumption that you want the channel values themselves to be NaN, rather than setting the Alpha value to be NaN for the dead pixels.

[Kevin Crosby]

Thanks for the prompt response. I'm trying to implement this now.

[Kevin Crosby]

I changed Photometric to MinIsBlack, since I'm dealing with grayscale images. You are correct that the pixels values themselves are NaN.

Assuming that I overwrite my original file in variable "filename" with the transformed image in "timg", here's the relevant code ...

info = imfinfo(filename);
t = Tiff(filename, 'w');
tagstruct.ImageLength = size(timg, 1);
tagstruct.ImageWidth = size(timg, 2);
tagstruct.Compression = Tiff.Compression.None;
tagstruct.SampleFormat = Tiff.SampleFormat.IEEEFP;
tagstruct.Photometric = Tiff.Photometric.MinIsBlack;
tagstruct.BitsPerSample = info.BitsPerSample; % 32;
tagstruct.SamplesPerPixel = info.SamplesPerPixel; % 1;
tagstruct.PlanarConfiguration = Tiff.PlanarConfiguration.Chunky;
t.setTag(tagstruct);
t.write(timg);
t.close();

Thanks again!

[Chris Volpe]

I've tried this, but it's converting my 32-bit float data to 8 BPP grayscale for some reason. My data is range data, specified in millimeters as a single precision float. I've tried both MinIsBlack and LinearRaw for the Photometric option, but no luck. Any thoughts?

Here's my code:

    t = Tiff(fullfile(filepath, [basename '.tif']), 'w');
    tagstruct.ImageLength = size(depthImage, 1);
    tagstruct.ImageWidth = size(depthImage, 2);
    tagstruct.Compression = Tiff.Compression.None;
    tagstruct.SampleFormat = Tiff.SampleFormat.IEEEFP;
    tagstruct.Photometric = Tiff.Photometric.LinearRaw;
    tagstruct.BitsPerSample = 32;
    tagstruct.SamplesPerPixel = 1;
    tagstruct.PlanarConfiguration = Tiff.PlanarConfiguration.Chunky;
    t.setTag(tagstruct);
    t.write(single(depthImage));
    t.close();

[Walter Roberson]

I just re-tested the code I posted at https://www.mathworks.com/matlabcentral/answers/307519-image-file-saved-using-tiff-library-tif-appear-white-in-explorer-and-paintbrush#answer_239772 and confirm that it creates single precision on my Mac in R2019a. The only difference I can see is the Photometric.