Add a chapter to handbook documentation about image data structures

[nopria]

It would be very useful to know, for each image mode, how image data is structured. For some kind of images it is quite obvious, but for others, especially if transparency is present, it is not (or at least I was not able to find it).

For example, I searched in documentation an explanation to this code (in order to check if it's correct and to adapt it to my needs) but I could find only sparse and partial information.

Thanks in advance for your attention.

[radarhere]

I imagine you've seen https://pillow.readthedocs.io/en/stable/handbook/concepts.html. I would hope that explains that RGBA is made up of four channels ranging from 0 to 255, where the fourth channel is alpha.

I've created PR #6831 to add a "Transparency" section to the documentation - https://pillow--6831.org.readthedocs.build/en/6831/handbook/concepts.html#transparency

Let us know if that is sufficient, or there are other items that are unclear.

[nopria]

Sure, answered beyond my most optimistic hope!

And I still think that the downgrade function above would be a useful core command...

[radarhere]

Why are you interested in this functionality? At first I thought maybe you were trying to minimise the output of tobytes() for storage purposes, but if you were trying to optimise storage, you'd be better off using an image format like PNG.

[nopria]

Since I have to produce PDF 1.3 (for printing purposes) with all kind of different images (photos, drawings, etc.) and the images must be not less than 300 DPI of real printing size, but also not much more than 300 DPI (to keep the PDF -- usually hundreds of pages -- as small as possible), my workflow force me to:

• downsample (seldomly upscale) each image at the number of pixels necessary to obtain just not less than 300 DPI in printing
• convert to sRGB colorspace and remove transparencies (PDF 1.3 does not support transparent images)
• identify almost greyscale images (usually color scan of greyscale originals) and force them to greyscale, which they benefit from
• quantize to about 32 colors images like drawings without graduated shadings, which quality often benefits from strong quantization
• convert all image formats to either JPG (photos) or PNG (any image with less than 256 colors)

In such a mess I just find very useful to have a way to convert image mode losslessly (i.e. without altering image content, optimizing the data structure of image if downgrading) just before saving it to JPG or PNG.

This downgrade function might be added to Pillow convert command easily, just adding a force parameter which would default to True, and users like me who want to convert to a specific image mode without altering image content could set force=False.

[radarhere]

Not all image formats support all modes when saving, so I'm concerned that converting to a lower mode might actually introduce an error in some circumstances. Perhaps if #2663 is solved, then that would provide a good basis for some sort of general optimize argument when saving.

This downgrade function might be added to Pillow convert command easily, just adding a force parameter which would default to True, and users like me who want to convert to a specific image mode without altering image content could set force=False.

Although here you seem to not be talking about a method to downgrade the mode as much as possible, but instead altering convert() so that if told to only be lossless, and the image isn't simple enough for that... it just returns the image unchanged?

[nopria]

Although here you seem to not be talking about a method to downgrade the mode as much as possible, but instead altering convert() so that if told to only be lossless, and the image isn't simple enough for that... it just returns the image unchanged?

Yes, I admit my explanation wasn't clear, because in my mind your downgrade function is the above expectations objective of a lossless mode convert command, i.e. a conversion affecting only image data structure leaving intact image content.