What’s your “image rehab” routine?

[SBT]

Given that a epub file must be less than 20MB (in the P. Clark Library, at least), how should one prioritize for image-rich books? Image size, jpeg quality, or colour depth?
Personally, I reduce colour depth for prints and b/w sketches, and quality for photos, paintings and coloured sketches. However, I'm more unsure about what dimensions to use. I tend towards a width of 800 px as default.
And how much priority should I give to minimizing the file size, versus keeping the images nice for my future retina-display e-reader?
Finally, what about prints and similar where moiree patterns are an issue; should I make the images no wider than 600px, so that they seldom will be downscaled, or should future-proof big images have priority?

[mrmikel]

I think the only safe way is to preserve your original files together in a folder and produce what is needed at the time.

Some devices have hi-res displays, but their users don't buy books for them so much. If you keep the largest original files, you can pretty easily knock out low res ones in an epub, rename the high res ones and add them to the epub or use a program that can do bulk resizing or bulk change in resolution for something in between.

Ebooks are hot right now, but whether they last for another ten years, who knows?

[SBT]

Originally Posted by mrmikel

Ebooks are hot right now, but whether they last for another ten years, who knows?

Never mind e-books - will I last another ten years?
Hopefully I will be thumbing through my old, well-worn, stained and crumpled e-books a dozen years from now

Ten years is as close to infinite time as to make no difference when it comes to gadgets, but I would like to make books that don't look utterly antiquated in let's say two years time.

Maybe the good people in the Calibre team could add an image-shrink functionality, for those books which are just a little too future-proof?

[Tex2002ans]

Originally Posted by SBT

Ten years is as close to infinite time as to make no difference when it comes to gadgets, but I would like to make books that don't look utterly antiquated in let's say two years time.

I agree... which is why I lean more towards higher resolution images in the books (I mean, we should abandon that crappy < 600px x 800px, it is pathetically small on these higher resolution devices).

I say make the source image as close to original as possible (do all your editing THERE), then shrink down if needed. A smaller/medium resolution image can always be generated from the source... although maybe I just don't know what I am doing. (I have zero art skills).

I mostly deal with charts/graphs/tables though, so my life is easier than dealing with illustrations/photographs. I can always recreate these as vector art, so I can scale to whatever size is needed.

[mrmikel]

Converting to vector image might be an ideal solution for tables. But conversion programs are not so common or inexpensive. Re-creation is a lot of work.

[Tex2002ans]

Originally Posted by mrmikel

Converting to vector image might be an ideal solution for tables. But conversion programs are not so common or inexpensive. Re-creation is a lot of work.

Well Tables can for the most part be recreated in HTML (although for my SVG Tutorial I AM planning on potentially tackling vector tables... an SVG Table MIGHT scale nicer (or be flipped at a 90 degree angle), I still have to do lots of testing and learn a lot more about SVG).

As for charts and graphs, here are a few graphs I was able to recreate using Inkscape in one night (again, I have zero vector art experience):

Page 103:



Page 104:



Page 261:



Page 262:



Side Note: Forgive for the "off-topicness", but I feel this is also an important area of "image rehab".

[Freeshadow]

I'm no math guy, but isn't feeding the equation resulting in a graph into an appropriate program (e.g. gnuplot) and exporting it to SVG from there better than drawing the stuff manually?

avogadro (molecular editor) can export to SVG and PNG too

[GrannyGrump]

@derangedhermit: I've downloaded CQ, and will be trying it out this weekend. I have to check on those resizing algorithms, I don't know what most of them do!

@SBT: How many second chances to chase the future do any of us have? . I'm going to be redoing Huckleberry Finn, Roughing It, and others changing from 600px wide to 800px wide (or maybe a bit wider, haven't quite decided). Most of the Twain books I have done are *very* image heavy, and book size will skyrocket, so I'm probably going to have to split them into two volumes to stay under MR's upload limit.
I don't know if it is the best thing to do, but I have usually set a max-width in pixels on my illustrations so they can't display more than 150% of the native size. They might be too small for hi-res displays, but at least that keeps them from upscaling so huge that they look totally ugly.

@freeshadow: I will take another look at RIOT. I tried it once a long time back, but my PSP got files sizes as good or better, so I set it aside. Maybe the newer version does better.

@Tex: I wish there was a way to keep my images at original size, but the file-size is just too intimidating for my purposes. I doubt that anyone would want to download a public domain book with 100 - 350 png illustrations weighing 1 - 5 MB each. (most are around 2500 px to 3500 px wide). Now if anyone ever invents a breed of svg that does "photo-real" stuff...

[Tex2002ans]

Originally Posted by Freeshadow

I'm no math guy, but isn't feeding the equation resulting in a graph into an appropriate program (e.g. gnuplot) and exporting it to SVG from there better than drawing the stuff manually?

No idea yet, that was just a few nights of fiddling a while back. I tried to use a nice amount of "automated" PNG -> SVG conversion tools, but most gave EXTREMELY bloated and subpar output.

For example, Inkscape has a "Tracing" tool:

http://inkscape.org/doc/tracing/tutorial-tracing.html

Instead of giving a single straight line, it would set many "points" along the line, and created a "jagged" very thin rectangle, or instead of a smooth curve line, it would generate hundreds of points along the line. Then using "simplify":

http://inkscape.org/doc/advanced/tutorial-advanced.html

to try to reduce the number of points just lead to breakage.

Perhaps I was just using the wrong tools (or maybe had to read much more closely in the documentation to get proper output), I will check out this gnuplot, thank you for the information.

Originally Posted by GrannyGrump

I don't know if it is the best thing to do, but I have usually set a max-width in pixels on my illustrations so they can't display more than 150% of the native size. They might be too small for hi-res displays, but at least that keeps them from upscaling so huge that they look totally ugly.

Indeed, this is a good idea. I would rather just have a smaller (slightly expanded) image, than a HIDEOUSLY upscaled image.

Originally Posted by GrannyGrump

@Tex: I wish there was a way to keep my images at original size, but the file-size is just too intimidating for my purposes. I doubt that anyone would want to download a public domain book with 100 - 350 png illustrations weighing 1 - 5 MB each.

I meant you keep/edit the original source, but putting smaller images into the EPUB (but larger than 600x800). Yes yes, a 300 MB EPUB would be immense.

Originally Posted by GrannyGrump

Now if anyone ever invents a breed of svg that does "photo-real" stuff...

Well that is up to the person who designed the original file... there is lots of good vector art out there, although it mostly lays in the realm of comics/digital drawings. Sadly, I don't think you will be seeing century-old Mark Twain illustrations as SVG.

[Jellby]

Originally Posted by Freeshadow

I'm no math guy, but isn't feeding the equation resulting in a graph into an appropriate program (e.g. gnuplot) and exporting it to SVG from there better than drawing the stuff manually?

For that kind of diagram, I like using a kind of "programming language" for graphics, such as the TikZ package for LaTeX

Originally Posted by GrannyGrump

I'm probably going to have to split them into two volumes to stay under MR's upload limit.

The limit might change when MR moves to the new software (no date for that yet, but we all hope it will be soon). Anyway, instead of splitting the book in two volumes, you could simply split the file, as I did with Don Quijote. The user still has to download two separate files, but after uncompressing there will be a single ePub.

[SBT]

Originally Posted by Tex2002ans

Sadly, I don't think you will be seeing century-old Mark Twain illustrations as SVG.

Hmm, sketches and etches should be translatable into a vector space, though... Individual strokes in a sketch are normally identifiable (unlike brush strokes), and presumably can be approximated by a limited number of parameters defining the properties of the stroke... And quite some research has already gone into stroke recognition as part of OCR on handwriting.
Could be an interesting research project...

Until then, the only time I vectorize images is when they have clearly defined silhouettes, and even then I only do it if there's a cover image I want to look extra nice.

[mrmikel]

I was dreaming of vectorizing both tables and the numeric and alphabetic content.

[Toxaris]

Originally Posted by mrmikel

I was dreaming of vectorizing both tables and the numeric and alphabetic content.

Then you woke up and found out you had a nightmare?

[GrannyGrump]

@Jellby --- split/spanned archives. Hmmm. That is an interesting idea. But I don't think Windows' built-in zip utility handles those, and not everybody keeps a third-party tool to handle rar, 7z, etc. (Let me give a recommendation for ArcThemAll, available from http://arcthemall.sourceforge.net/. Its Free and Open Source, and a pretty nice little tool that can unpack dozens of archive formats, including rar, 7z, zip, tar/gz, etc, and can archive to zip or 7z.) If Windows can handle split archives, I would be interested to know how to do that, because it surely would be useful.