Re: Bug with B&W conversion

Hi Bob,
I've just spent some time reprocessing some of my sample images with the Local Adaptive Thresholding with different parameters.  The results are good in that they expose a lot of extra details but unfortunately they are also increasing the noise level and are not suitable for the project I'm currently doing as they are aesthetically less appealing by a large margin.  That said I will definitely be using the LAT method on other projects where OCR is more important as the results of LAT are definitely more OCR-friendly than the dithering/error-diffusion of the Riemersma routine.
Given that the Riemersma dither is default, and basically any B&W conversion (explicit or implicit) done by ImageMagick/GraphicsMagick will use this routine, shouldn't this be a priority to get fixed.  By my estimates 2% of my images suffer from this bug and are left blank.
Regards,
Phillip

> Date: Sun, 9 Jan 2011 19:24:03 -0600
> From: [email protected]
> To: [email protected]
> CC: [email protected]
> Subject: Re: [GM-bugs] Bug with B&W conversion
> 
> On Mon, 10 Jan 2011, Andrew Stonewall wrote:
> 
> > Hi Bob,
> > This bug goes back to the ImageMagick base-code for the 'riemersma' dither routine.  I raised this on the
> > ImageMagick forums some time back (http://www.imagemagick.org/discourse-server/viewtopic.php?f=3&t=17251) but
> > had a hard time convincing their admin that it was a real problem.
> 
> There is definitely a weakness in the algorithm, but it may be 
> difficult to solve.  Color reduction is really intended for full-range 
> photos.
> 
> > The main reason I'm using ImageMagick (and now GraphicsMagick) is 
> > that this routine provides an exceptional B&W conversion on some 
> > very 'noisy' documents that I'm processing at the moment so using 
> > another dither isn't an option.
> 
> Yes, the quality of the dither can be very nice.  Slow but nice.
> 
> Both -ordered-dither and -random-threshold provide recognizable output 
> from your image but the results are not as nice.
> 
> An alternative is to use local adaptive thresholding (-lat).  This 
> algorithm is good for bringing out the original features of the scan. 
> The best parameters to use depend on your input image and require some 
> experimentation.
> 
> Here is an example:
> 
>    gm convert original.bmp -colorspace gray -lat 10x10-4% output.bmp
> 
> In this example, it computes the average value of the pixels in a 
> 10x10 rectangle around the target pixel.  Then it thresholds based on 
> the pixel value being more than 4% different than the average.  This 
> algorithm can use a lot of CPU so it is best to keep the rectangle 
> size down to the smallest size which reliably works.
> 
> The end result captures all of the written/detail bits, while 
> eliminating the page background.  Perhaps this is what you wanted in 
> the first place?
> 
> Bob
> -- 
> Bob Friesenhahn
> [email protected], http://www.simplesystems.org/users/bfriesen/
> GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/

------------------------------------------------------------------------------
Gaining the trust of online customers is vital for the success of any company
that requires sensitive data to be transmitted over the Web.   Learn how to 
best implement a security strategy that keeps consumers' information secure 
and instills the confidence they need to proceed with transactions.
http://p.sf.net/sfu/oracle-sfdevnl

_______________________________________________
Graphicsmagick-bugs mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/graphicsmagick-bugs