How to prevent the saved images become uint8

Dear Admin,

This is my code:

filename = 'D:\files.txt';
grayImage = transpose(importdata(filename));  
a = readmatrix(filename);                          
imwrite(a', 'files.png');

I want to read a 2D array (512x512 with double type) from a matrix file named ‘files.txt’. However, when I try to save this matrix as an image, the data type of the image becomes uint8.

My problem is how to preserve the original data type (double) from ‘files.txt’ when saving it as an image?

Regards,

[Dyuman Joshi]

That's not possible.

The default behaviour when using imwrite is to convert the bit depth to uint8. Exceptions include some image types that can be saved as uint16.

As your file is a 2D array of double type, I assume it is a grayscale image.

From the documentation of imwrite - "If A is a grayscale or RGB color image of data type double or single, then imwrite assumes that the dynamic range is [0, 1] and automatically scales the data by 255 before writing it to the file as 8-bit values."

[Niko]

Dear Dyuman,

Thank you for your explanation. I appreciate your insight into the default behavior of imwrite and its conversion to 8-bit values. I had previously read the documentation and was aware of this behavior.

However, I was wondering if there might be a way to save the image as a double type in JPEG or PNG format. I understand from your explanation that this might not be possible due to the limitations of these formats, but I thought I’d ask just in case there might be a workaround or alternative method. Thank you again for your assistance.

Regards,

Answer 1

[Walter Roberson]

PNG has no support for floating point pixels.

TIFF has support for floating point, but the imwrite() interface to TIFF does not support that.

See https://www.mathworks.com/matlabcentral/answers/307519-image-file-saved-using-tiff-library-tif-appear-white-in-explorer-and-paintbrush#answer_239772 for sample direct MATLAB code

See https://www.mathworks.com/matlabcentral/fileexchange/30519-export-image-to-tif-or-tiff-file-of-selected-data-type for a File Exchange contribution that knows how to deal with the various different TIFF output types. (Except possibly COMPLEX)

[Niko]

Thank you for your prompt response, Walter.

I have successfully saved the image in double datatype using the imwrite2tif function from the link you provided. However, I noticed that I cannot open the saved image using Windows Photo Viewer, and it displays a message stating that “Photos can’t open this file because the format is currently unsupported, or the file is corrupted”. I suspect this might be due to the image’s double datatype, as I was able to open an image saved in uint8 datatype normally.

You mentioned that “PNG has no support for floating point pixels”. Does this also apply to JPEG? In other words, does JPEG also not support floating point pixels?

Lastly, are there any default functions in MATLAB, other than imwrite2tif, that can save an image in double datatype?

Thank you again for your assistance.

[Image Analyst]

You can use save to save the variable perfectly but it's mostly just understandable by MATLAB. I don't believe JPEG (https://en.wikipedia.org/wiki/JPEG ) can save floating point pixel values.

You should tell us where (what software) you want to use the saved image and how. Not every program and operating system out there understands all potential image file formats. So you need to know exactly what your plans are for using the saved file. Tell us where/how you will use it.

[Walter Roberson]

The work-around for PNG files is to take the 64 bit double data and typecast() it to 'uint8' and reshape to an array with 8 times as many rows as previous. PNG is lossless so it will write the exact bytes out. It will look wrong on viewers, and you will need to reverse the conversion when you read it back in from the png file, so it is not clear why you would want to do this... you might as well just write it as a binary file if you were going to do this.

The work-around for JPG is similar to the above, but when you imwrite() it you must specifically tell imwrite() to use lossless JPEG.

The code I posted in https://www.mathworks.com/matlabcentral/answers/307519-image-file-saved-using-tiff-library-tif-appear-white-in-explorer-and-paintbrush#answer_239772 shows how to use the built-in Tiff class to write TIFF with floating point.

geotiffwrite can write floating point layers (to TIFF files), if I recall correctly.

To be explicit:

• bmp, gif, jpg / jpeg, jp2, pbm, pcx, pgm, png, pnm, ppm, ras, xwd -- none of these can support floating point pixels at all (other than by converting the
• tif / tiff can support floating point pixels but imwrite() does not support that for tiff
• hdf file format can support floating point, but the built-in image types for hdf do not support floating point
• some image formats such as JPEG commonly support EXIF metadata and by using "private" tags you could write an unstructured binary blob that only your program would know how to deal with. However, each blob has a maximum size that makes it inconvenient for larger images, and there is no support in any Mathworks-provided function (that I know of) that can write EXIF.
• PNG file format also supports private tags that can hold arbitrary binary data, but finding code that can write such tags is more difficult; no Mathworks-provided function (that I know of) handles this.
• Additional image file formats exist (not supported by MATLAB) that can store floating point pixels. In particular, there are some formats specifically designed for HDR (High Dynamic Range) that support floating point pixels.

I believe I have heard that the utility ImageJ is able to render floating point TIFF.

Floating point TIFF is a real part of an international standard, not a hack. It just isn't handled by the default Windows viewer. Floating point tiff is used extensively by Mapping and GIS systems, but those typically need some interpretation to render into images.

[DGM]

I think it's worth emphasizing that the solution depends less on MATLAB and more on whatever other application you expect to read the file. Most image viewers and editors (at least that I've tried) won't open a floating-point TIFF (or at least not correctly).

Similarly, you might be able to typecast the data and pack it into a PNG for transport, but most other applications will have no way of reversing the process, and any modifications performed on the packed image will destroy it.

If the target application supports no formats that are capable of floating-point, then that's that. If there is no target application, and all you are concerned with is using it in MATLAB, then why not use a MAT file or something? A MAT file will be reasonably compact and convenient to use -- at least within MATLAB.

[Niko]

Image Analyst: Thank you for your response. I have tried the save function and also looked into the JPEG file format on Wikipedia. The reason I want to use floating point pixel values is because I believe it will provide more precise data than the uint8 pixel values. I plan to import these images into Lumerical Ansys, which only accepts .jpg or .png image types. Unfortunately, .tiff files cannot be imported into Lumerical Ansys. I understand from your response that JPEG may not support floating point pixel values, but I’m wondering if there’s a workaround or an alternative format that might be compatible with Lumerical Ansys.

Walter Roberson: I appreciate your detailed explanation and the code you provided. I was able to successfully generate a .tiff image with float double data. Your information regarding which image file formats support floating point was very helpful. Your suggestion about typecasting the 64-bit double data to ‘uint8’ for PNG files is interesting, and I’m considering giving it a try. However, I’m concerned about the potential loss of precision during the conversion process. Do you think this could be a significant issue?

DGM: Thank you for your insights. Like you, I have tried opening images with double point pixel values in various applications without success. My goal is to use these float files in Lumerical Ansys. Your suggestion about using a MAT file within MATLAB is a good one, but unfortunately, it doesn’t solve my problem with Lumerical Ansys. Do you have any other suggestions for formats or methods that might be compatible with this software?