LibGfx/JPEG: Support for progressive JPEG with huge end-of-bands run
Support for progressive JPEGs in LibGfx is brought by #17907.
However, it doesn't have support for at least one edge case: images with huge end-of-bands. Indeed, a normal EOB marker can only embed a value of, at maximum, 32 767. But some images might need to encode more than that.
As specified in G.1.2.2 Progressive encoding of AC coefficients with Huffman coding:
If an EOB run is greater than 32 767, it is coded as a sequence of EOB runs of length 32 767 followed by a final EOB run sufficient to complete the run.
This is something that happens to work on baseline JPEGs and SOF2 with spectral selection only. In other words, it only fails with progressive JPEGs.
Required equipment
- A copy of the repo, specifically both
add_acandread_eobfunctions inJPEGLoader.cpp - A copy of the JPEG specification (a link can be found in
JPEGLoader.h) - A test image
(which is included, don't worry), hmm actually my method to generate one didn't work, so you will have to create one.
Test image
Small explanation on how I generated the image:
I created a 1500x1500 uniform image in GIMP and exported it in .ppm (a format that cjpeg can read), then I created the final image with cjpeg -progressive.
Now, why I did that:
- We need to generate EOBs which are added to the stream when multiple following blocks (or bands inside blocks) contains only zeros.
- To ensure that zeros are present in a band, the MCU needs to have small variations of values around the mean. Hence, a uniform color for the whole image ensure us to have AC components with a value of zero, so perfect conditions for a lot of EOB.
- Finally, we want at least 32 768 blocks in a scan, so the image needs to be big enough to include that much MCUs. A small computation shows that 1500 / 8 > 187 which gives a total of a minimum number of MCUs of 187*187 = 34969 > 32768.
Edit:
My bad, this method doesn't work as cjpeg seems to hit another condition before reaching an EOB of 32768. A few other things to try:
- Use a bigger image as an input
- Tweak
cjpegcompilation options - Use another encoder
Recommended equipment
- Discord to ask me questions 😃
I added the good first issue label, as this issue can probably be fixed in less than 10 lines. However, it probably requires a bit more than common knowledge (read that as you will need to read the spec) to understand what you're doing.
How does funding with Polar work?
Pay now to fund the work behind this issue.
Get updates on progress being made.
Maintainer is rewarded once the issue is completed.
Backer
You're funding impactful open source efforts
Contributor
You want to contribute to this effort
Maintainer
You want to get funding like this too