Closed GoogleCodeExporter closed 8 years ago
The algorithm we use is the Median cut, but it's done in plain RGB colorspace
instead of some perceptive one. And there are some hacks that were meant to
make it work better, but in the end make it actually worse on some pictures.
Also, the current implementation of the Median cut needs a lot of memory
(several megabytes) for various tables. Likely something can be done to improve
that.
Original comment by pulkoma...@gmail.com
on 23 May 2011 at 8:19
I just spent half an hour uselessly trying to relive the old algorithm
(comparing the french and english source code line by line), I give up for
tonight. Anyway, if we have defective color perception, we should at least try
to do it in 128 times less memory. And a practical goal would be the support of
transparent RGB images : a specific RGB triplet designates the transparent
color, we need to resolve a palette slot for it and reduce the others to 255.
It exists in RGB and PNG, IIRC.
Original comment by yrizoud
on 23 May 2011 at 10:34
DawnBringer, can you provide the original truecolor image? I tried
color-reducing the "Photoshop" part of your example, and strangely all the 256
colors were preserved. Yeah I was about as tired as when I wrote the above
comment (128->64, resolve->reserve, RGB and PNG -> *BMP* and PNG...)
Original comment by yrizoud
on 14 Jun 2011 at 10:52
Here...
Original comment by annas...@hotmail.com
on 14 Jun 2011 at 11:41
Attachments:
fixed in r1878.
Original comment by pulkoma...@gmail.com
on 28 Nov 2011 at 10:31
Original comment by pulkoma...@gmail.com
on 8 Jan 2012 at 3:39
Original issue reported on code.google.com by
annas...@hotmail.com
on 23 May 2011 at 8:13Attachments: