Does this library support writing jpg files with only 1 or 2 channels to a pdf?

[ChloeOfAstora]

I am writing a simple file conversion program where I have a bunch of images in a folder and turn those into a pdf at some point, I just made a function to enable converting the images to greyscale only, and while the file size of the images is a lot lower, the pdf only shrinks a couple kb in size on a 42mb pdf.
Is this because there is no support for using images with fewer channels in this library (or pdf's in general) or am I just using the wrong function for this?
Right now i'm using the pdf_add_image_file function.

[AndreRenaud]

The JPEG images basically get copied in unchanged, since PDF supports JPEG decoding natively. So you should see a file size reduction. Can you send me an example of the problem (ie: the jpeg files etc...)?

[ChloeOfAstora]

here is one example image in both greyscale and colour,

and here are the pdf sizes for the 2 versions with 300+ colour/grayscale images

maybe I just messed up somewhere else in my code but I was expecting a bigger difference.

[AndreRenaud]

Those numbers don't seem too surprising to me - the greyscale version is only 9kB smaller (7%). The final PDF is ~6% smaller. So that roughly tracks. The surprising thing is that making the jpeg greyscale in the first place didn't save much space.

Using the images you sent through, I was able to get slightly better compression (16%) using ImageMagick - how are you converting them from colour -> grayscale?

I used:
convert colour.jpg -colorspace Gray grey.jpg

The initial size of the image you sent was 130750, your grayscale one was 121602 (7% reduction), imagemagick's grayscale one is 110109 (16% reduction).

[ChloeOfAstora]

yes, my code is at https://github.com/ChloeOfAstora/cbz2pdf
I do save it to a jpg file with compression with stb_image_write.h

[AndreRenaud]

I suspect that stbi_write_jpg doesn't do as good of a job at compressing things as other JPEG libraries. It is designed for small code size/simplicity, but not bulk performance. Could you do the conversion outside your c++ code using imagmagick? I see you're basically unpacking a .zip file, and converting each one to grayscale, before deleting it. If you don't care about the speed of this, you could even just do system("convert " + tempPath.string() + "/" + entry.path().filename().string() + " " + filenameNew) (ignore the terrible string handling).

[ChloeOfAstora]

Yes I could do that, before this I was using a bash script that I wrote to do basically the same thing, just using a lot of external utilities like imagmagick, I wanted to try making a program that would just be one file and work everywhere. So if I want to keep it internal in the program I guess I'll have to look for a better performance jpg library.

[AndreRenaud]

Yeah, that makes sense. Having it all nicely self-contained is a good feature (I mean, that's the whole point of this PDF library).
Unfortunately there's not much that PDFGen can do in this case - apart from parsing the header to work out the width/height, the reset of the JPEG data is just copied unchanged into the PDF.

[ChloeOfAstora]

Thank you very much for your help and fast responses!
And also thank you very much for your amazing work with this library.

[AndreRenaud]

Thanks. I'm glad you're getting some use out of it. Good luck with the jpeg compression stuff.