>>121348I've been reading the specification for the PNG file format so I'm going to derail this thread a little bit here
Let's do some math: an uncompressed is mainly an array of pixels, let's assume the pixels have 4 channels (RGBA) and each channel takes one byte, that's 4 bytes per pixel. In reality image files also contain more info like metadata, headers and so on, but these are usually negligible in size.
so:
>>121289has
2894 * 4093 = 11845142 pixels
which amount to
(11845142 * 4) / 1000000 = 47.38 Megabytes, that means the png has a compression ratio of
47.38 / 10.72 = 4.42
Pretty impressive if you ask me, the image was compressed to 23% of its original size, but this is actually not good enough for a png
Compare this to
>>121343which has
1130 * 1684 = 1902920 pixels
which amount to
(1902920 * 4) / 1000000 = 7.61 Megabytes
The compression ratio here is 34.22, or we could say that the image was compressed to %3 of its original size, the reason here is mainly because this image contains a lot of the same color (white) which is handled pretty well by run length encoding
And looking
>>121289 I assumed at least there would be better compression on that shab's face and hair, but upon further inspection, it appears this image contains more gradients than it looks like, which explains the bad compression ratio here
This can be seen more clearly by "uncorrecting" the gamma, i did this by performing a linear invert, followed by a regular color invert with gamma correction. Surprisingly the image size has been lowered, I assume this is because the roof was blackened? PNG filters are indeed complicated and work in mysterious ways