Ximagic Quantizer

[eL_PuSHeR]

I have a question for you. I have decreased colour depth to 256 colours using binary split + fan and binary split + Ulichney. Both images ended up with the same size. Why is that?

Even more: doing a binary comparison between the two file shows they are byte-by-byte identical.

[brodriguez]

I´ve tested the two conversions that you mention, and the result are different for my test images.

Can you send me your image ?

[eL_PuSHeR]

That's weird. I have used pngout (by Ken Silverman) afterwards. It shouldn't matter. This is the image.

[brodriguez]

I've tested the quantizations binarysplit+fan and binarysplit+ulichney and I get different result files with different size.

The RMSE of the two transformations are different:
Binary Split + Ulichney : 4.887203
Binary Split + Fan : 4.866917

With PSP the sizes of the output files are:
BinarySplit + Ulichney : 216.099
BinarySplit + Fan : 215.525

Comparing the output files with WinMerge, the are different in nearly 100% of the content.

Could be a problem in the plugin host ?
Which image editor are you using ?

[eL_PuSHeR]

No. I have found the reason: I FEEL LIKE A MORON NOW...

I had Quantization box checked but neither Threshold dithering nor Error dithering checked. The plugin should warn you about this (idiot-proof).

Thanks for your efforts.

[brodriguez]

Don't feel so, I think it is a UI fault, it has too much controls and don't disable it according to selections.

There is a new version available (v 2.42) with some UI changes:
Now the controls are disabled or enabled according to the "Threshold Dithering"/"Quantization"/"Error dithering" check buttons.
The preview window is cleared when configuration changes.

I'm sure you can't make the same mistake with this UI

[eL_PuSHeR]

Thank you very much, friend.

[eL_PuSHeR]

A small update just to say that I have tested the latest versions (denoiser, quantizer, graydither) and they work flawlessly.

[brodriguez]

Thanks for your update.

I'm now working to solve 2 issues.
A.- The plugins don't allow to cancel processing. This can be a real pain sometimes, for instance in the scolorq quantization, the DBS dithering or the NLMeans denoising.
B.- When Quantizer or GrayDither exits because it can't allocate enough memory, sometimes not all the memory allocated by the plugin is released.

When I finish with those issues I'll try to make the plugins multicore friendly.

[eL_PuSHeR]

Thanks. Yesterday I try gray-dithering a big image 9300x13000 I think and it couldn't allocate enough memory. Solution: I sliced the image in two halves.

[brodriguez]

This is something the plugin must do for you, but I know it doesn't.

Anyway, you should be able to process this image without splitting it.

In the last versions, I make huge image test with GrayDither using a grayscale test image with 12288 x 12288 pixels and those are the results with several editors:
PS - OK
PSP - OK
Gimp - OK
IrfanView - Not enough memory
Photobie - Crash
PhotoPerfect - Crash
XNView - OK

The image you mention has only 80% of the size of my test image, so it should work, but nothing is sure when you are using a computer.

I make the tests starting the editor and making only the minimal operations, load the image and a apply the filter. This is important when you are at the memory limits, the memory is in it best shape when a program starts and usually gets worse as you use the program.

Always take into account the undo buffer. If you modify an image, the editor need to save the old version to be able to allow undos. Some editors allow to disable the undo (f.i. IV) or has options to delete the undo buffer (f.i. PS).

If the image you use is a color one, then the plugin and the editor will use more memory, a lot more. Try converting it to grayscale, save it to file , close the editor, open it again and use GrayDither with the grayscale image.

Do those explanations help you

[eL_PuSHeR]

I will try what you say. Yes, working with truecolour images needs at least three times more memory.

Thank you very much.

[eL_PuSHeR]

Update:

I have tested the new installers (MSI) and MP support for Denoiser. It seems to work just fine under both Single-Core and Multi-Core microprocessors. Tested on an AMD Sempron 64bit and AMD Phenom Quad-Core 64bit.

[brodriguez]

I have tested the new installers (MSI) and MP support for Denoiser. It seems to work just fine under both Single-Core and Multi-Core microprocessors. Tested on an AMD Sempron 64bit and AMD Phenom Quad-Core 64bit.

I'm a bit worried about using msi installers, it is something I prefer to avoid, but in this case is the best solution.

So I'm very happy to receive confirmations that the installer works.

The MP version should work theoretically as the SP version when there are only one logical processor (nor multicore nor hyperthreading) but it is good to have tests which confirm that.

I expect to publish the MP versions of Quantizer and GrayDither in three days. MP will make some algorithms, like DBS dithering or ckmeans quantization, more usable.

Thank you very much for your testing and feedback.

PD: Days ago I discovered one use of b/w dithering I didn't know about. Someone opened a thread about XiGrayDither in a woodworkers forum http://www.sawmillcreek.org/showthread.php?t=114467. It seems laser engravers use B/W images as input, at least the one they mention in the forum.

[eL_PuSHeR]

Ximagic Graydither is very useful for sending output to B/W laser printers.