1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
|
- in light of twitter, improve palette (aka PNG8) support.
https://twitter.com/NolanOBrien/status/1080914181061111808
- need a tool to change bitdepth, pridepthpng
- pipcolours (needs renaming) should also count colours and
create palette and suggested palette chunks.
use JASC-PAL format? see
https://sourceforge.net/projects/jascpal2png/ ?
http://www.selapa.net/swatches/colors/fileformats.php is quite
a good overview
- Should a paletted PNG with a grey palette come out grey? (as
NetPBM's pngtopnm tool seems to do) Probably yes? So that grey
-> palette -> direct comes out as grey
- priweavepng should have a magic palette mode. See below.
output palette image.
- texttopng should take a font PNG. such as https://github.com/drj11/pypng/issues/57
- when converting rows (asRGB, asDirect, and so on), could pass
in an info structure with desired settings (eg
info['bitdepth'] = 8 would return 8-bit values;
info['greyscale'] = False would return RGB (or RGBA);
info['colormap'] = False would convert palette image)
# priweavepng magic palette mode
if all selected channels are from
the same paletted image and
priweavepng should output a colour-indexed image,
with tRNS chunk if applicable.
# Low bit-depth palettes
Obervation:
- palette entries are 8-bit colour
16-bit PNGs should not be paletted.
Colour PNGs are always 8- or 16-bit.
Palette entries should be drawn from 8-bit pixel entries.
sBIT chunk should be transferred.
# Reducing numbers of colors
When reducing the numbers of colors,
is it possible to separate the reduction of colors from
assigning palette members to full spectrum colors?
I think it might not be.
Consider reducing many greyscales to 4:
xx x x x x xx
12 4 29 31 96 98 99
Indeed, the following picks 4 colours, but then only uses 3 of them:
printf 'P2 8 1 100\n1 2 4 29 31 96 98 99\n' | pnmquant 4 | pnmtoplainpnm
The problem here is that 4 of Heckbert's boxes are formed, but
two of them only have elements near their extreme edges.
The boxes are reduced to a colour near their centre,
but that colour is not the nearest box centre to any of the
original colours.
This suggests that it may not be simple to separate the steps:
- choosing a palette;
- mapping colours to palette members.
One solution may be to record the boxes of the Heckbert median
cut algorithm.
An axis aligned box can be specified by any two (body) diagonal corners.
The output of this step can be a 2 by N PNG image.
Note that having the intermediary be a sequence of boxes means
that the palette element isn't represented in the intermediary.
This is acceptable if the palette element is a function of the box
(for example, it is the centre of the box).
This is unacceptable if the palette element is chosen in other ways
(for example, the most popular input colour).
A 3 x N image could record a box and a palette element.
|
