Compression of Canon CR2 raw images in MATLAB ?

I am using my Canon 5D Mark IV to acquire images of star fields. Often in Astro photography the most interesting features have very low intensity compared to the noise level.

The noise level in RAW camera images should not be exposed to compression schemes. I enclose one image above, where the red channel of the image displayed (an astronomical image with a few high intensity stars) is a correct representation of what MATLAB displays - but clearly there is a problem with the lowest intensity levels of the image. It appears as if MATLAB is utilizing lossy compression scheme, which in the appeararance looks like JPEG compression.

I would like to know how to switch off the compression, if anybody knows.

Here is the code used to display the image in MATLAB

[myfile,mypath]=uigetfile({'*.cr2'});
>> mygray2=myim(:,:,1);
>> mytiny2=mygray2(2000:2500,3300:3800);
>> figure(3);
>> imshow(mytiny2,[10 30]);

Below is shown the same image - but processed with PIXINSIGHT - please note the noise background in this image - no lossy compression artefacts, no large square patches- a telltale of compression.

[Walter Roberson]

You show selecting the file, but not reading it?

[Steffen B. Petersen]

You are correct - somehow that line was not copied, mea culpa, sorry:

myim=imread([mypath '\' myfile]);

[Walter Roberson]

I would question whether it is a matter of "compression" or if instead the reading is at fault.

I would probably try calling upon https://pypi.org/project/rawpy/ to see if I could get a better image.

[Steffen B. Petersen]

I am very happy working inside the MATLB environment - and I have already shown that Pixinsight does a better job on the background (no indication of compression).

Also it is somewhat of a jungle concerning the different raw formats. Canons raw format for the 5D mark iv eg is not the same as the 5D mark iii.

[Image Analyst]

imread() does not read .cr2 files. Type imformats to see what files can be read in.

So, since you can't be using imread(), what is the line of code you're using to read in the .cr2 files?

Also I've heard that cr2 files are not really raw images. I think I heard that they already have dead pixel correction applied, and they may also have demosaicing applied.

[Walter Roberson]

I do see evidence that people have been able to use imread with cr2 but most of them seem to be using octave or at least Linux.

[Walter Roberson]

Steffan, is it your hypothesis that the "compression" is being done by imshow, or is your hypothesis that the "compression" is being done by the rendering? There is some antialiasing of fonts going on, and you can control about alignment of logic pixels against physical pixels, but those are things that would affect all images.

In the case of rendering an image at higher resolution than it physically is then you can get block artifacts that would show up for every thing. In the case of rendering an image at lower resolution than it physically is then there can be interpolation.

My hypothesis is that the imread of the cr2 is pulling in a substandard quality image rather than there being a problem with MATLAB "compression".

[Steffen B. Petersen]

Hi Walter

The problem is still not solved. I spoke to MATLAB and they state that their imread and imshow does not introduce compression. Also spoke to Canon support. They were not helpful at all - basically stating that their CR2 format was secret and they brushed me off, stating that I bought a normal consumer camera and I could not expect any special services.

The Canon 5D mark 4 camera produces beatiful images of size 30 MP - so I have plenty of resolution.

Reading the image header file, the CR2 file appear to consist of 4 TIFF files of different sizes and compression codes. One TIFF file seem to be empty, size is 0 - and maybe this is where to dig for the real raw data.

[Walter Roberson]

Try adding an index to the imread call. Beyond that I would try with Tiff()

Answer 1

[Bray Falls]

For sanity purposes I'm letting you know I also experience severe compression effects when loading CR2 files into MATLAB. I am processing solar eclipse images, and after filtering the images they are full of square tiles and other artifacts. Converting to .tiff in pixinsight is fine but I'd like to read the raw file directly.