Re: [Gimp-user] Size of output files

[Ofnuts <ofnuts laposte net>]

On 27/02/15 14:03, Alan Barnett wrote:
> I'm using gimp to increase the contrast of scanned documents.  The 
> scanned images were combined into a multi-page pdf.  The images are 
> music scores; the originals are RGB, but only grayscale information is 
> important.
> I opened the input pdf as layers with a resolution of 300 dpi, changed 
> the brightness and contrast of each layer, and exported the images as 
> a .mng file.  I then used ImageMagick to convert the nmg back to .pdf 
> format.
>
> I've done this with two different sets of images.  For the first set 
> of images, the input .pdf files are about 2 Mb, and the output .mng 
> and .pdf are both about 4 Mb; the process increases the size of the 
> files by a factor of 2.
>
> For the second set of images, the input pdf files are about 5 MB, and 
> the output .mng and .pdf files are both about 45 MB, an increase of a 
> factor of 9!
>
> Any suggestions how to decrease the size of the output files? Any 
> ideas about why the large difference in the size increase between the 
> two sets of images.  (The first set were scanned by me on an EPSON 
> 1640SU scanner, the second set were scanned by someone else on a 
> scanner unknown to me.)


The file size is related to the size of the image in pixels, not to its 
physical size (which depends on print defintion)

Also, in the PNG format (and AFAIK MNG is a PNG derivative) there is no 
compression of random data, so color uniformity is important to keep 
size to a minimum.

An album of three pictures(*) on IMGUR:

http://imgur.com/a/a3ZRb

- The one where the background is pure white is 4K.
- Rather visible solid noise added to the clean image increases it to 20K
- Barely visible HSV noise added to the clean image increases it to 30K

So some pos-tprocessing to cleanup your background may have the required 
effect. Thresholding would be perfect for the final size but would be 
ugly. High contrast could do it, or a more fine-tuned Curves setting. A 
de-speckling filter could also help.


(*) Anyone recognizes the song?